# paper.pdf ## Metadata - PDFFormatVersion=1.6 - IsLinearized=true - IsAcroFormPresent=false - IsXFAPresent=false - IsCollectionPresent=false - IsSignaturesPresent=false - CreationDate=D:20260512093327+05'30' - ModDate=D:20260512093350+05'30' - Producer=Acrobat Distiller 10.0.0 (Windows) - pdf:keywords="transparency"; "translucency"; "perception"; "IT cortex"; "macaque monkey"; "neuronal selectivity" - dc:format=application/pdf - dc:title=Interaction of optical parameters in the perception of transparency/translucency and their neural representation in the visual cortex - dc:identifier=doi:10.1167/jov.26.5.4 - dc:description=Journal of Vision, 2026. doi:10.1167/jov.26.5.4 - prism:doi=10.1167/jov.26.5.4 - prism:url=https://doi.org/10.1167/jov.26.5.4 - prism:aggregationtype=journal - prism:publicationname=Journal of Vision - prism:issn=1534-7362 - dc:creator=Mika Baba,Kei Kanari,Kei Iwasaki,Hidehiko Komatsu - dc:subject=transparency,translucency,perception,IT cortex,macaque monkey,neuronal selectivity - xmpmm:documentid=uuid:d224de3b-be58-3bbf-a1b5-da42f5e760c2 - xmpmm:instanceid=uuid:23d16dad-cd48-3828-8c09-26f1efb85500 ## Contents ### Page 1 Journal of Vision (2026) 26(5):4, 1–26 1Interaction of optical parameters in the perception of transparency/translucency and their neural representation in the visual cortex Mika Baba Brain Science Institute, Tamagawa University, Machida, Tokyo, Japan RIKEN, CBS, Wako, Saitama, Japan National Institute for Physiological Sciences, Okazaki, Aichi, JapanKei Kanari Brain Science Institute, Tamagawa University, Machida, Tokyo, Japan Advanced Institute of So-Go-Chi Informatics, Tohoku University, Sendai, Miyagi, JapanKei Iwasaki Graduate School of Science and Engineering, Saitama University, Saitama, Saitama, JapanHidehiko Komatsu Brain Science Institute, Tamagawa University, Machida, Tokyo, Japan National Institute for Physiological Sciences, Okazaki, Aichi, Japan National Institute for Basic Biology, Okazaki, Aichi, JapanTransparency and translucency are important visual attributes that serve as cues for object recognition and discrimination. The interaction between light and objects involves complex processes encompassing multiple factors. In this study, we systematically manipulated parameters of light transmission, scattering, and reflection to investigate their effects on perceived transparency/translucency in human subjects. We found that the extinction coefficient, a parameter governing transmission, exerts the greatest influence on perceived transparency/translucency across objects of varying shape complexity. However, albedo, which indicates the amount of scattering, also significantly affects the perceived transparency/translucency of objects with relatively simple geometries. Additionally, surface roughness, a reflection parameter, strongly influenced perceived transparency/translucency. We then examined single-neuron activities in the inferior temporal cortex of the macaque monkey, where previous studies have identified gloss-selective neurons. We recorded neural responses using stimulus sets comprising the same parameters tested in our psychophysical experiments. Many neurons exhibited selectivity for parameters related to perceived transparency/translucency, including extinction coefficient, albedo, and surface roughness. Furthermore, population responses of neurons selective for these parameters showed interaction patterns consistent with those observed in our psychophysical experiments. This suggests that neurons selective for these parameters in this area are related to the perception of transparency/translucency. IntroductionMany natural and artificial objects transmit light to varying degrees. The extent of light transmission depends on material properties and the object’s scale, resulting in diverse object appearances ranging from highly transparent glass to human skin (which transmits light partially) to stone surfaces (whichCitation: Baba, M., Kanari, K., Iwasaki, K., & Komatsu, H. (2026). Interaction of optical parameters in the percep- tion of transparency/translucency and their neural representation in the visual cortex. Journal of Vision, 26(5):4, 1–26, https://doi.org/10.1167/jov.26.5.4.https://doi.org/10.1167/jov.26.5.4 Received September 18, 2025; published May 12, 2026 ISSN 1534-7362 Copyright 2026 The Authors This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. ### Page 2 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 2transmit minimal light). Transparency refers to the state in which light passes through the material with little scattering and objects behind the material can be seen. On the other hand, translucency refers to the state in which light passes through the material, but it is scattered, and objects behind the material appear blurred and unclear. Transparency and translucency constitute important visual attributes, and differences in these properties serve as critical cues for object discrimination. The question of which visual cues underlie transparency/translucency perception has attracted the interest of many researchers. Early transparency studies aimed to determine how overlapping surfaces could be distinguished based on color and brightness cues, using stimuli with artificial geometric shapes and patterns (Beck & Ivry, 1988; Khang & Zaidi, 2002; Metelli, 1974). However, advances in computer graphics (CG) technology have enabled accurate simulation of light–material interactions and intraobject light behavior. Consequently, recent studies have extensively used images of transparent or translucent objects with complex three-dimensional (3D) shapes (Chowdhury, Marlow, & Kim, 2017; Fleming, Jensen, & Bülthoff, 2004; Gkioulekas et al., 2013; Marlow, Kim, & Anderson, 2017; Nagai et al., 2013; Sawayama et al., 2022; Xiao et al., 2014). In a pioneering work, Fleming and Bülthoff (2005) suggested that various mid-level image features relate to translucency perception. While contrast and blur have been shown to be related to the perception of transparency in filter models (Singh & Anderson, 2002a; Singh & Anderson, 2002b), Motoyoshi (2010) demonstrated that the relationship between the contrast and blur of specular highlights and nonspecular areas is related to the perception of translucency in 3D objects. While individual differences exist regarding which part of an object is focused when perceiving translucency (Nagai et al., 2013), the structure of an object’s surface (such as 3D shape and scale) influences the perception of translucency. Fleming and Bülthoff (2005) demonstrated the importance of edges in highly light-transmitting objects, and Gkioulekas, Walter, Adelson, Bala, and Zickler (2015) demonstrated that the luminance profile of edges carries information related to translucency. Thinner parts of objects (Gkioulekas et al., 2013; Gigilashvili, Thomas, Hardeberg, & Pedersen, 2020) and rugged surfaces (Sawayama et al., 2022) tend to yield stronger perceived transparency. Illumination direction also significantly influences the perception of translucency (Fleming & Bülthoff, 2005; Xiao et al., 2014). Under forward illumination, the luminance gradient due to subsurface scattering covaries with the shape of self-occluding contours and specular reflections in translucent objects, suggesting that these cues provide both material and shape information (Marlow & Anderson, 2021). Despite these and many other studies (see Gigilashvili, 2021b, for review), however, it is still not fully understood what image cues are involved in the perception of transparency/translucency. Key physical parameters affecting light transmission through objects include the absorption coefficient (σ a ), which indicates the degree of light absorption per unit distance, and the scattering coefficient (σ s ), which indicates the extent of light scattering and directional change within the object. The extinction coefficient (σ t = σ a + σ s ), representing the sum of these variables, indicates the degree to which light intensity decreases over a given distance and has been shown to affect object transparency/translucency (Xiao et al., 2014). Additionally, psychophysical experiments have demonstrated that the phase function—a physical parameter indicating light-scattering directionality— also affects transparency/translucency perception, with strength of the effect depending on illumination direction (Gkioulekas et al., 2013; Xiao et al., 2014). However, perceived translucency of object is also influenced by specular highlights that is related to the property of surface light reflection (Fleming & Bülthoff, 2005). In this study, we systematically manipulated parameters of light transmission, scattering, and reflection across wide ranges to investigate their effects on perceived transparency/translucency. In our first test, we varied the extinction coefficient over a wide range and combined it with various albedo values to examine how object transparency/translucency is affected by absorption and scattering. This test used object images with a specific phase function and smooth surfaces. In subsequent experiments, we varied the phase function and surface roughness while maintaining the same extinction coefficient and albedo ranges, examining these parameters’ effects on perceived transparency/translucency. We employed state-of-the-art CG techniques to create photorealistic object images with systematically varied parameters, asking participants to rate image transparency to determine which physical parameters contribute to transparency/translucency perception. Results showed that the extinction coefficient exerts the greatest effect on perceived transparency/translucency across objects of varying shape complexity, while albedo also affects perceived transparency/translucency in objects with relatively simple shapes. Phase function effects were relatively small but significant, confirming previous reports (Gkioulekas et al., 2013; Xiao et al., 2014). Surface roughness also strongly affected perceived transparency/translucency, providing quantitative support for previous observations (Fleming & Bülthoff, 2005). In our psychophysical experiments, participants ### Page 3 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 3were asked to evaluate the perceived transparency of objects. However, our stimuli encompassed a wide range of absorption and scattering parameters, spanning very transparent to translucent to opaque materials. Additionally, previous transparency studies often employed thin, see-through filter-like stimuli, which differ substantially from our stimuli. For these reasons, we use the term transparency/translucency in this article’s title and preceding text to avoid misunderstandings. However, to avoid redundancy in subsequent descriptions, we will use transparency unless otherwise necessary. Compared to psychophysical studies of trans- parency, little research has examined the neural basis of transparency/translucency perception. Here, we investigated whether neurons in macaque monkey visual cortex—whose perceptual functions resemble those of humans—show selectivity for object transparency using images in which physical parameters related to transparency (identified in our psychophysical exper- iments) were systematically manipulated. Regarding light reflection, electrophysiological studies in macaque monkeys and functional magnetic resonance imaging (fMRI) experiments in humans and macaques have identified gloss-selective activity in higher areas of the ventral visual pathway (Nishio, Goda, & Komatsu, 2012; Okazawa, Goda, & Komatsu, 2012; Sun, Ban, Di Luca, & Welchman, 2015; Wada, Sakano, & Ando, 2014). In our electrophysiological experiment, we recorded from the central inferior temporal (IT) cortex, where gloss-selective neurons have been previously observed (Baba, Nishio, & Komatsu, 2021; Nishio et al., 2012). To examine the relationship between transparency and gloss selectivity at the neuronal level, we tested neuronal selectivity using stimuli combining physical parameters related to transparency and glossi- ness. For transparency, we manipulated the extinction coefficient and albedo; for glossiness, we manipulated the object surface roughness. Our results demonstrate that IT cortex contains neurons selective for physical parameters related to transparency, with some neurons showing selectivity for both transparency- and glossiness-related parameters. Furthermore, population responses of parameter-selective neurons exhibited interaction patterns consistent with those observed in our psychophysical experiments. These findings are the first to show how transparency/translucency- related information is represented in the brain. The rest of this article consists of two parts. In the first part (Experiment 1), the procedures and results of the psychophysical experiments will be described, followed by the discussion of this part. In the second part (Experiment 2), the electrophysiological experiments will be described. At the end of the article, a general discussion will be made.Experiment 1—Psychophysics in human subjects MethodsVisual stimuliA set of object images was created as stimuli using the Mitsuba renderer (Jacob, 2010) to manipulate a wide range of physical parameters related to the transmission, absorption, and scattering of light within an object. First, 3D shapes with three levels of complexity were generated. The complexity levels ranged from a smooth ellipsoid-like shape (shape S2), a shape of intermediate complexity (shape S7), and a complex bumpy shape (shape S11) (Figure 1 for S7; Supplementary Figure S1 for S2 and S11). These shapes were parametrically generated using an algorithm based on spherical harmonics functions (Shimokawa, Nishio, Sato, Kawato, & Komatsu, 2019). The shape9 14 18 22 25 7 12 17 21 24 5 10 15 19 23 3 6 11 16 20 1 2 4 8 13 0.90.1 0.3 0.5 0.7 1.6 0.8 0.4 0.2 0.1 albedo extinction coefficientFigure 1. An example of a stimulus set used for psychophysical experiments to examine the perceived transparency of object images. The stimulus set consists of 25 object images, and each object had a different combination of one of five levels of the extinction coefficient (σ t ) (0.1, 0.2, 0.4, 0.8, 1.6) and five levels of albedo (σ s /σ t ) (0.1, 0.3, 0.5, 0.7, 0.9). Object shape is S7 for this set. Each combination of parameters was numbered 1 to 25 for convenience, as indicated in the top-left corner (not shown in the stimulus). The number corresponds to the ascending order of the scattering coefficient (σ s ) of the rendered object. All 25 combinations were used in the main experiment, but only 9 combinations indicated by a yellow frame and an underbar were used for psychophysical experiments to examine the effects of phase function and roughness. Object images for shapes S2 and S11 are shown in Supplementary Figures S1A and S1B, respectively. ### Page 4 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 4index numbers (S2, S7, S11) represent the upper limit of the degree of the spherical harmonic functions that are randomly combined, with larger numbers corresponding to increased bumpiness. In the basic stimulus set, objects were rendered with a total of 25 different parameter combinations, combining five levels of the extinction coefficient (σ t= σ s + σ a )—a parameter indicating how quickly light attenuates within an object—and the albedo (σ s /σ t )—a parameter indicating the degree of scattering within an object (Figure 1). Five levels of the extinction coefficient σ t were selected at equal intervals on a logarithmic axis: 0.1, 0.2, 0.4, 0.8, and 1.6. Five levels of albedo were selected at equal intervals on a linear axis: 0.1, 0.3, 0.5, 0.7, and 0.9. We employed σ t and albedo that are wavelength-independent, resulting in mostly grayscale appearance of objects. We chose to use a stimulus set combining extinction coefficient and albedo to create object images with various appearances and transparencies. These parameters have also been used in previous studies to describe light behavior within objects (Gigilashvili, Shi, et al., 2021; Xiao et al., 2014). In the present study, we determined the range of each parameter that can generate large variation of object appearance based on a pilot test and prepared object image sets by systematically combining these parameters within the ranges. The directional distribution of light scattering is described by the phase function. We employed phase functions defined on the two-dimensional (2D) plane determined by computational analysis and psychophysics (Gkioulekas et al., 2013) and used the same identification numbers to specify phase functions as Xiao et al. (2014). For the basic stimulus set combining five extinction coefficients and five albedo values, we employed phase function #12. In the basic stimulus set, objects had smooth surfaces with a surface roughness value (Beckmann α) of 0.01. For this and other stimulus sets, we used a natural illumination environment (Eucalyptus Grove) from the Debevec dataset (https://www.pauldebevec.com/Probes/). The objects were approximately 4 cm in size and placed on a dark gray plane. The interior of each object was modeled as a uniform scattering medium with a refractive index of 1.5. After rendering, the high dynamic range images were gamma-corrected based on sRGB gamma, tone-mapped to 8-bit color images, and then converted to grayscale using the rgb2gray function in MATLAB (MathWorks). The final image size was 256 × 256 pixels. In addition to the basic stimulus set, we created image sets with four different phase functions (phase-function stimulus set) and two different levels of surface roughness (surface-roughness stimulus set). First, from the 25 basic stimulus combinations mentioned above, we selected nine parameter combinations (three extinction coefficients, 0.1, 0.4, and 1.6, and three albedo values, 0.1, 0.5, and 0.9) that coarsely sampled the extinction coefficient and albedo across the same range (selected stimulus set: Figure 1, yellow frame). For the phase-function stimulus set, we created 9 ×4 = 36 images in total, applying four different phase functions (#1, #4, #9, #12) to each of the nine images from the selected stimulus set (Supplementary Figure S2). Of the 16 phase functions used by Xiao et al. (2014), we selected numbers 1, 4, 9, and 12 because they are maximally separated in the 2D parameter space. The surface roughness value was 0.01 for the phase-function stimulus set. For the surface-roughness stimulus set, we created 9 × 2 = 18 images in total, applying two surface roughness values (0.01, 0.1) to each of the nine images from the selected stimulus set (Supplementary Figure S3). Objects with α = 0.01 had smooth surfaces, while those with α = 0.1 had rougher surfaces. Phase function #12 was used for the surface-roughness stimulus set.ProcedureTo investigate how parameters related to light behavior within an object affect perceived transparency/translucency, we conducted a psychophysical experiment in which naive human participants rated the perceived transparency of computer-generated object images. We recruited participants with normal eyesight over 1.0, which was self-reported. The experiment was conducted in a dark room using images presented on a liquid crystal monitor (EIZO CS2420-ZBK). The viewing distance was 57 cm, and the stimulus size was about 6° in visual angle. The monitor had a gamma of 2.17 and 256 gradation levels for each RGB color channel. The background luminance was 0.6 cd/m 2 , and the average luminance of the stimulus images ranged from approximately 2 to 22 cd/m 2 . Visual stimulus presentation and experimental control were managed on a MacBook Pro (macOS Sierra, 10.12.6; Apple) using Psychtoolbox software (Brainard, 1997; Pelli, 1997) running on MATLAB (MathWorks). Participants viewed each image presented on the monitor in a dark room and responded by entering their perceived transparency rating using number keys from 0 to 99, where 0 represented completely opaque and 99 represented completely transparent. Instruction was given in Japanese without any specific definition oftransparency. Stimuli were presented until participants responded. The presentation order was randomized, and each stimulus was presented three times to each participant. The average of the three responses was used for subsequent analysis. First, participants rated the perceived transparency of each of the 25 images in the basic stimulus set. The three object shapes (S2, S7, S11) for the basic stimulus set were tested in separate blocks, with the presentation order of shapes randomized across participants. Ten participants completed this test (four males and six females, ages 19–25). The participants took about ### Page 5 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 50.1 0.3 0.5 0.7 0.9 albedo 0.1 0.2 0.4 0.8 1.6 extinction coefficient shape S2 60 50 40 30 200.1 0.3 0.5 0.7 0.9 albedo 0.1 0.2 0.4 0.8 1.6 shape S7 60 50 40 30 20 700.1 0.3 0.5 0.7 0.9 albedo 0.1 0.2 0.4 0.8 1.6 90 80 70 60 50 40 30 20 10 0 perceived transparency shape S11 60 50 40 30 20 70 10A B CFigure 2. Results of the experiments using the stimulus set depicted in Figure 1 (for shape S7) and Supplementary Figure S1 (for shapes S2 and S11) to examine the effects of the extinction coefficient (σ t ) and albedo (σ s /σ t ) on the perceived transparency of objects. (A) Average ratings of the perceived transparency (0–99) of the stimuli with shape S2 for nine participants are plotted as different shades of pink in a matrix format. The vertical axis of the matrix represents the extinction coefficient (σ t ), and the horizontal axis represents the albedo (σ s /σ t ). Black lines indicate contours connecting the points where the magnitude of the ratings is the same. (B) Average ratings of the stimuli with shape S7. (C) Average ratings of the stimuli with shape S11. The same data in the graph format plotted separately against the extinction coefficient or albedo with SD are shown in Appendix A.10 min to access 75 stimuli (25 stimuli × 3 times) on average, so it required about 30 min to complete the ratings for three shapes. We then conducted an experiment in which participants rated the perceived transparency of the 36 phase-function stimuli to examine the effect of phase functions on perceived transparency. We also conducted an experiment with the 18 surface-roughness stimuli to examine the effect of surface roughness on perceived transparency. The phase function and surface roughness tests used the object shape of intermediate complexity (shape S7). Twelve participants completed the phase-function and surface-roughness tests (seven males and five females, ages 19–22). Participants were recruited through a platform within Tamagawa University in which students willing to participate in psychophysical or fMRI experiments are preregistered. They were paid for their participation in the experiment in the present study. The experiments conformed to the Declaration of Helsinki, and participants provided written informed consent before the experiments. In all tests, the experimenters were not included among the participants. The human participant experiments were conducted with approval from the Tamagawa University Ethics Committee (TRE19-0005s).ResultsEffects of σ t and albedo on perceived transparencyTen participants evaluated the perceived transparency of each image in the basic stimulus set across three shapes generated by combining various σ t and albedo values. One participant showed results with an opposite pattern compared to other participants and was excluded from analysis because this participant appeared to have misunderstood the response instructions (Supplementary Figures S4.1 and S4.2). The averaged results from the remaining nine participants are shown in Figure 2 in a matrix format (A for shape S2, B for S7, C for S11). For all shapes, perceived transparency decreased significantly as σ t increased along the vertical axis. The same data are shown in graph format in Appendix A. Although the magnitude of perceived transparency tended to be lower for the simplest shape, perceived transparency decreased almost linearly with the logarithm of σ t for all shapes, except at the smallest σ t values. This result is consistent with Xiao et al. (2014), indicating that σ t controls perceived transparency. Simultaneously, we found that albedo had a clear effect on perceived transparency. For the complex shape (S11), iso-response contours are roughly parallel with the horizontal axis, but for the simpler shapes (S2, S7), iso-response contours have larger slopes against the horizontal axis, indicating that perceived transparency varied with albedo. Such a tendency can also be clearly seen in Figure A1 in Appendix A, in which perceived transparency (vertical axis) is plotted against albedo (horizontal axis). For shapes S2 and S7, perceived transparency tended to decrease as albedo increased, whereas for shape S11, the curves were nearly parallel to the horizontal axis with little change across albedo values. Standard deviation (SD) of the ratings shown in Figure A1 in Appendix A is fairly large, suggesting that there are interparticipant variations in the responses. We quantitatively evaluated the interobserver and intraobserver agreements of the ### Page 6 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 6ratings (Supplementary Material 4). There were modest interparticipant correlations (Pearson’s correlation coefficients: average 0.56–0.70; Supplementary Table S1) and intraparticipant correlations (Pearson’s correlation coefficients: average 0.60–0.75; Supplementary Table S2). We conducted a two-way repeated measures analysis of variance (ANOVA) to examine the influence of σ t and albedo on perceived transparency. Significant main effects of σ t were observed for all shapes (S2: F(4, 32) =16.69, p < 0.001; S7: F(4, 32) = 36.07, p < 0.001; S11:F(4, 32) = 29.36, p < 0.001). For albedo, significant main effects were observed for S2 (F(4, 32) = 14.68, p< 0.001) and S7 (F(4, 32) = 9.31, p < 0.001), but no significant effect was observed for S11 (F(4, 32) = 1.72,p > 0.05). While no significant interaction was observed between σ t and albedo for S2 and S11 (p > 0.05; S2:F(16, 128) = 1.31; S11: F(16, 128) = 0.63), there was a significant interaction for S7 (F(16, 128) = 3.67, p <0.01). Therefore, for shape S7, we conducted simple effects analysis with Bonferroni correction for multiple comparisons to examine the effect of albedo at each level of σ t . The effect of albedo was significant at σ t= 0.2 and 0.4, but not at other levels (for σ t values of 0.1, 0.2, 0.4, 0.8, 1.6: F(4, 32) = 2.24, n.s.; F(4, 32) =9.93, p < 0.001; F(4, 32) = 12.59, p < 0.001; F(4, 32) =3.26, n.s.; F(4, 32) = 0.64, n.s., respectively, where n.s. indicates p > 0.05). These results are consistent with the observations described above.Effect of phase function on perceived transparencyNext, to examine how changes in phase function affect perceived transparency, we investigated the effect on perceived transparency using object images created with the same range of σ t and albedo values as above, rendered with four different phase functions (phase-function stimulus set). Twelve participants evaluated the perceived transparency of images created with one object shape (S7). One participant showed results with an opposite pattern compared to other participants and was excluded from analysis because this participant appeared to have misunderstood the response instructions. Figure 3 shows four example images (A) and the experimental results (B). The images in the upper row show the combination of σ t = 0.4 and albedo = 0.5 (Figure 1: stimulus number 15), and the images in the lower row show the combination of σ t =1.6 and albedo = 0.9 (Figure 1: stimulus number 25), both rendered with four different phase functions. To examine the influence of phase function on perceived transparency, we performed a two-way repeated-measures ANOVA with phase function and stimulus number as factors. We found that both factors showed significant main effects (phase function:F(3, 30) = 17.69, p < 0.001; stimulus number: F(8, 80) = 54.48, p < 0.001), and there was a significant#1 #4 #9 #12 phase function 100 80 60 40 20 0 perceived transparency 9 18 25 5 15 23 1 4 13 stimulus number albedo=0.9 albedo=0.5 albedo=0.1 t =0.1 0.4 1.6 0.1 0.4 1.6 0.1 0.4 1.6 #9 #12 phase function #1 #4 A Bstim #25#15Figure 3. Results of the experiment to examine the effects of phase function on the perceived transparency of object images. (A) Example images of stimuli number 15 (top row) and 25 (bottom row) rendered with four phase functions (#1, #4, #9, #12). Stimulus numbers correspond to those depicted in Figure 1. All images used for this test are shown in Supplementary Figure S2. (B) Average ratings of the perceived transparency (0–99) for 11 participants (vertical axis) are plotted against the stimulus number. Actual values of extinction coefficient (σ t ) and albedo (σ s /σ t ) for each stimulus number are shown below. Lines with different colors correspond to different phase functions. Error bars indicate SD.interaction between the two factors (F(24, 360) = 3.22,p < 0.001). We conducted simple effects analysis with Bonferroni correction for multiple comparisons to examine the effect of phase function at each stimulus number. The effect of phase function was significant at stimuli #15 (F(3, 30) = 12.98, p < 0.001), #18 (F(3, 30) = 7.44, p < 0.01), and #25 (F(3, 30) = 6.83, p< 0.05), but not at other stimulus numbers (F(3, 30) ### Page 7 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 7< 5.12, p > 0.05). The result that phase function affects perceived transparency is consistent with previous studies (Gkioulekas et al., 2013; Xiao et al., 2014). The three stimuli with significant effects (#15, #18, #25) are relatively high albedo materials. This seems reasonable because for low albedo materials that have low scattering, the impact of the phase function indicating light-scattering directionality should be low.Effect of surface roughness on perceived transparencySurface roughness is a major parameter affecting the perceived gloss of an object, but Fleming and Bülthoff (2005) noted that highlights also contribute to the visual impression of translucency. This suggests that the surface roughness of objects also affects perceived transparency. To quantitatively examine this issue, we tested the effect of surface roughness on perceived transparency using images rendered with two different surface roughness values—a smooth surface (α = 0.01) and a rougher surface (α = 0.1)—for object images created with the same range of σ t and albedo values as above (surface-roughness stimulus set). The same 12 participants from the phase-function test evaluated the perceived transparency of images created with one object shape (S7). One participant— the same one excluded from the phase-function test—showed results with an opposite pattern compared to other participants and was excluded from analysis. Figure 4 shows two example images (A) and the results (B). The left and right images show objects rendered with two roughness values (left: 0.01; right: 0.1) and a combination of σ t = 0.1 and albedo =0.9 (Figure 1: stimulus number 9). The object on the left appears to have a smooth surface with sharp highlights, while the object on the right has a frosted glass-like appearance with blurred highlights. Results in Figure 4B show a common tendency for perceived transparency to decrease as σ t increases for both αvalues. We note that the results for α = 0.01 (black line) are very similar to those shown in Figure 3, indicating good reproducibility. When comparing results for the two α values, the magnitudes of perceived transparency differ substantially, with the smooth-surfaced object (black line) being perceived as more transparent. To examine the influence of surface roughness on perceived transparency, we performed a two-way repeated-measures ANOVA with roughness and stimulus number as factors. We found that both factors showed significant main effects (roughness: F(1, 10) =56.62, p < 0.001; stimulus number: F(8, 80) = 48.86,p < 0.001). There was also a significant interaction between the two factors (F(8, 80) = 17.98, p < 0.001). We conducted simple effects analysis with Bonferroni correction for multiple comparisons to examine the effect of roughness at each stimulus number. The effect100 80 60 40 20 0 perceived transparency 9 18 25 5 15 23 1 4 13 stimulus number albedo=0.9 albedo=0.5 albedo=0.1 t =0.1 0.4 1.6 0.1 0.4 1.6 0.1 0.4 1.6 roughness 0.01 0.1 stimulus number: 9 roughness: 0.01 roughness: 0.1 A BFigure 4. Results of the experiment to examine the effects of surface roughness (α) on the perceived transparency of object images. (A) Example images of stimulus number 9 rendered with two values of roughness (left; 0.01, right; 0.1). Stimulus number corresponds to that depicted in Figure 1. All images used for this test are shown in Supplementary Figure S3. (B) Average ratings of the perceived transparency (0–99) for 11 participants (vertical axis) are plotted against the stimulus number. Actual values of extinction coefficient (σ t ) and albedo (σ s /σ t ) for each stimulus number are shown below. Lines with different colors correspond to different roughness. Error bars indicate SD.of roughness was significant at stimuli #1 (F(1, 10) =30.13, p < 0.01), #4 (F(1, 10) = 95.55, p < 0.001), #5 (F(1, 10) = 51.44, p < 0.001), #9 (F(1, 10) = 30.13, p< 0.01), #15 (F(1, 10) = 23.34, p < 0.01), #18 (F(1, 10) = 20.47, p < 0.01), and #25 (F(1, 10) = 16.80, p< 0.05), but not at stimuli #13 and #23 (F(1, 10) <12.37, p > 0.05). This result indicates that, except for object images with large σ t values, surface roughness clearly affects perceived transparency. Therefore, even when the attenuation and scattering behavior of light within the object remains constant, differences in object surface smoothness affect perceived transparency, ### Page 8 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 8supporting previous findings (Fleming & Bülthoff, 2005; Gigilashvili, Dubouchet, Pedersen, & Hardeberg, 2020).DiscussionThe effect of albedo on the perceived transparencyIn the psychophysical experiments, we observed that albedo affected perceived transparency. For relatively simple object shapes, the perceived transparency of objects with the same σ t decreased when the albedo was large. Since albedo is σ s /σ t , this value increases when light scattering within the object is high and decreases when light absorption is high. The effect of albedo may be explained by the transparency of the background. For simple object shapes, the background is more visible through the object, and the blurring of the background introduced by high albedo (scattering) is easily noticed. In contrast, for complex shapes, complex surface geometry scatters light and obscures the background, and the additional blurring introduced by high albedo is less noticeable. When albedo is high, the amount of light that is not absorbed within the object exits in various directions from multiple locations on the object’s surface, increasing the object’s lightness. Increased lightness will reduce the contrast on the object surface, and this may be another possible reason why perceived transparency decreases when albedo is high. It has been shown that for highly transparent objects, perceived transparency increases with increasing RMS contrast (Motoyoshi, 2010, Figure 4). Additionally, as described below, perceived transparency has a strong relationship with object gloss. In translucent objects, Gigilashvili, Shi, et al. (2021) have shown that decreasing albedo in smooth spherical objects increased perceived glossiness. It is possible that the effect of albedo on perceived glossiness may indirectly influence perceived transparency. We observed that the influence of albedo on perceived transparency occurs when the object shape is relatively simple, with the effect being especially clear for the simplest shape (S2). Conversely, the effect was not observed for objects with a complex shape (S11). It has been reported that perceived transparency is related to cues about the object’s shape, which may be related to why the albedo effect was not seen in complex-shaped objects. Marlow et al. (2017) showed that the covariation between an object’s surface orientation and shading is related to perceived translucency. For objects with complex shapes, processing the covariance between shape cues and shading cues may be complicated, making it difficult for albedo to affect perceived transparency/translucency. There may be another reason why albedo did not affect perceived transparency for complex-shaped objects. Unlike opaque objects, bright spots on transparent object surfaces are created not only by highlights from reflected light but also by background light that has passed through the object’s interior. Gigilashvili, Shi, et al. (2021) noted that for objects with complex surface shapes, bright spots on the object surface are created by internal scattering when the albedo is high. For complex shapes, the increase in highlight contrast at low albedo and the increase in bright surface spots due to scattering at high albedo may counteract changes in glossiness, resulting in no change in perceived transparency due to albedo. We should note that for objects with intermediate complexity of shape (S7), the effect of albedo was significant for medium (0.2 and 0.4) σ t levels but not for very low (0.1) or very high ones (0.8 and 1.6). This indicates that the effect of albedo is less noticeable when the material is too opaque or when it has little scattering. The image features involved in the effect of albedo on perceived transparency/translucency require further research.Relationship between perceived transparency/ translucency and glossinessFleming and Bülthoff (2005) first demonstrated that object glossiness affects transparency/translucency, which has been confirmed by several other studies (e.g., Gigilashvili, Thomas, Hardeberg, & Pedersen, 2021). Motoyoshi (2010) showed that the relationship between highlights and shadows is an important cue for perceiving translucency/transparency. Regarding the relationship between object surface roughness—which greatly affects gloss—and transparency/translucency, Gigilashvili, Dubouchet, et al. (2020) varied the roughness of highly light-transmitting objects from 0 to 1 and showed that perceived translucency decreases monotonically with increasing roughness. The present study showed that increasing roughness generally reduces the perceived transparency of objects, not only for highly light-transmitting objects but also for objects with various absorption and scattering properties, except for objects with large σ t values. This supports previous findings that glossiness is linked to perceived transparency/translucency, although we have not explicitly examined the perceived glossiness of the stimuli in the present study. Fleming and Bülthoff (2005) suggested that although the optical processes related to glossiness and translucency are independent, many translucent objects (e.g., plastic, candle wax, or honey) possess some degree of glossiness. Therefore, the human visual system may expect translucent objects to be glossy (Fleming & Bülthoff, 2005, p. 353). If the visual system uses such an expectation, nonglossy objects may be more easily interpreted as opaque. Transparency and translucency are thought to reflect the optical properties of an object’s interior, not its surface. However, the ### Page 9 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 9results reported by Gigilashvili, Dubouchet, et al. (2020) and the present study suggest that humans may have a perceptual characteristic that misattributes the optical properties of the object’s surface to its interior. Specifically, when an object’s surface reflection is blurred, the object may be interpreted as having optical properties that do not transmit light well, even if the object material is highly light-transmissible (such as glass). Gerardin et al. (2019) proposed a 3D space model expressing transparency using three axes: absorption, internal scattering, and surface roughness. In our experiment’s stimulus set combining σ t and albedo, perceived transparency was mapped onto their absorption and internal scattering planes. By investigating the effect of surface roughness, we also explored the remaining dimension of their 3D space.Effect of phase function on perceived transparency/ translucencyGkioulekas et al. (2013) showed that translucent appearance varies depending on the phase function, which represents the spatial characteristics of scattering. The present study also showed that perceived transparency changes depending on the phase function, confirming their results, although the effect was observed within a limited range of σ t and albedo values. Xiao et al. (2014) noted that the phase-function effect is strong in thin objects, but in thick objects, higher-order scattering effects become dominant, and the phase-function effect weakens. The stimulus objects we used had relatively simple shapes with few thin sections, which may have made the phase-function effect relatively small in the present study. Even for thick objects, appearance changes depending on the phase function when front-lit (figure 5 in Gigilashvili, Thomas, et al., 2021). However, the lighting used in our experiment was an omnidirectional natural lighting environment with bright illumination mainly from behind, so such large appearance changes likely did not occur.Transparency and translucencyThe appearance of light-transmitting objects spans a wide spectrum depending on the degree of light absorption and scattering, changing from completely transparent to translucent to opaque. In our psychophysical experiment, we asked participants to evaluate the perceived transparency of object images. We believe this is a reasonable measure for evaluating this wide spectrum. One reason is that degree of translucency is an ambiguous term, and when asked to judge translucency, the meaning may differ between participants. It is ambiguous whether high translucencymeans a more transparent state or a state that is neither transparent nor opaque (Gigilashvili, Thomas, et al., 2020; Gigilashvili, Thomas, et al., 2021). We thought that asking about the degree of transparency would be less ambiguous and result in more consistent judgments between participants. Although there is a possibility that transparency and translucency represent different categories of perceptual appearance, we agree with the notion that “internal scattering gradually makes a perfectly transparent medium more translucent and eventually opaque. The boundary among them is fuzzy, implying that transparency and translucency are not mutually exclusive” (Gigilashvili, Thomas, et al., 2021, p. 6). As Motoyoshi (2010) noted, in Japanese, the word for transparent is tohmei, and the word for translucent is han-tohmei (han = half), so translucent is expressed as half-transparent. Unlike English, there is no independent word for translucency in Japanese. All participants in the present psychophysical experiment were Japanese speakers, and the instructions to evaluate transparency were likely natural for them. Whether there are perceptual differences between speakers of languages in which transparency and translucency are conceptually distinct and those in which they are not is an interesting question for future research, as it relates to the dimensions that underlie perceived transparency/translucency.Experiment 2—Electrophysiological recordings in macaque monkeyIn a previous study, we observed that gloss-selective neurons are concentrated in the lower bank of the superior temporal sulcus (STS) within the middle portion of the IT cortex of the macaque monkey (Baba et al., 2021; Nishio et al., 2012). In the present study, we recorded single-neuron activity from the same region while a monkey performed a fixation task and examined whether neurons in this area exhibit selectivity for parameters related to object transparency, such as light absorption and scattering, as well as selectivity for surface roughness, which is related to glossiness. To determine whether the activity of neurons selective for these parameters relates to transparency perception, we examined whether the neuronal responses exhibit properties analogous to the interparameter dependencies observed in psychophysical experiments.MethodsVisual stimuliThe three object shapes used for stimuli were identical to those in Experiment 1 (shapes S2, S7, S11). In Experiment 2, we prepared two stimulus sets ### Page 10 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 10albedo0.1 0.5 0.9 1.6 0.4 0.1 surface roughnessextinction coefficient 0.3 0.1 0.01 1.6 0.4 0.1 0.3 0.1 0.01 surface roughness 0.1 0.5 0.9 A B stimulus set a b a b randomized set t / albedo stimulus set t / albedo randomized set albedo extinction coefficientFigure 5. (A) (a) An example of a stimulus set (σ t –α stimulus set) used for an electrophysiological recording experiment to test the response selectivity for the extinction coefficient (σ t ) and surface roughness (α). The stimulus set consists of nine object images, and each object had a different combination of one of three levels of extinction coefficient (σ t ) (0.1, 0.4, 1.6) and three levels of roughness (α) (0.01, 0.1, 0.3). The albedo of all images is 0.5. (b) An example of a stimulus set (σ t –α randomized set) in which pixels were randomly rearranged within the object contour of each image in the σ t –α stimulus set. (B) (a) An example of a stimulus set (σ t –albedo stimulus set) used for an electrophysiological recording experiment to test the response selectivity for the extinction coefficient (σ t ) and albedo (σ s /σ t ). The stimulus set consists of nine object images, and each object had a different combination of one of three levels of the extinction coefficient (σ t ) (0.1, 0.4, 1.6) and three levels of the albedo (σ s /σ t ) (0.1, 0.5, 0.9). Roughness (α) of all images is 0.01. (b) An example of a stimulus set (σ t –albedo randomized set) in which pixels were randomly rearranged within the object contour of each image in the σ t –albedo stimulus set. Object shape is S7 for these four stimulus sets. Object images for shapes S2 and S11 are shown in Supplementary Figures S5 and S6, respectively.as visual stimuli for electrophysiological experiments in monkeys. The first stimulus set consisted of nine images combining three levels of extinction coefficient (σ t ) with three levels of surface roughness (Beckmannα). The extinction coefficients were identical to those in the selected stimulus set from Experiment 1, with three levels of 0.1, 0.4, and 1.6, while surface roughness had three levels of 0.3, 0.1, and 0.01. This stimulus set is called the σ t /α stimulus set (shape S7, Figure 5Aa; S2 and S11, Supplementary Figures S5Aa, S6Aa). Preliminary observations showed that 3D shape was not clear because the contrast on the object surface ### Page 11 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 11was too low when α = 1 was used for roughness, which maintains constant intervals with 0.01 and 0.1 on the logarithmic axis. Therefore, we used the smaller value of 0.3 instead of 1 for the three roughness levels. In a preliminary experiment conducted by the experimenters (HK, MB) as participants, we confirmed that using the three α levels mentioned above (0.3, 0.1, 0.01) maintained shape perception while ensuring roughly evenly spaced perceived transparency levels (Supplementary Figure S7). A previous study (Gerardin et al., 2019) also demonstrated the complex nonlinearity of the roughness scale. The second stimulus set consisted of nine images combining three levels of extinction coefficient (σ t ) with three levels of albedo (σ s /σ t ). The combinations of extinction coefficient and albedo were identical to those in the selected stimulus set from Experiment 1 (σ t : 0.1, 0.4, 1.6; albedo: 0.1, 0.5, 0.9). This set is called the σ t /albedo stimulus set (shape S7, Figure 5Ba; S2 and S11, Supplementary Figures S5Ba, S6Ba). Since the average luminance of each image differed in both sets, changes in neuronal response with stimuli might be attributable to luminance changes. Therefore, as control images, we created shuffled image sets in which pixels were randomly rearranged within the object contour and tested neuronal responses to these stimuli as well. These image sets are called the σ t /α randomized set and σ t /albedo randomized set, respectively (shape S7, Figures 5Ab, 5Bb; S2, Supplementary Figures S5Ab, S5Bb; S11, Supplementary Figures S6Ab, S6Bb). The rendering method for the above stimulus sets was identical to that in Experiment 1. After rendering, high dynamic range images were first converted to 8-bit RGB images using the tone-mapping method of Reinhard, Stark, Shirley, and Ferwerda (2002), and then converted to grayscale. During experiments, visual stimuli were presented using a graphics board (VSG2/3; Cambridge Research Systems). Visual stimuli were presented at approximately 9° of visual angle.Surgery and electrophysiological recordingsWe recorded neuronal activity from one hemisphere of a male macaque monkey (Macaca fuscata; weighing 6 kg). During experiments, the monkey was seated in a primate chair facing a 22-in. CRT monitor (TOTOKU, CV921X). Before starting experiments, a head holder and recording chamber (rectangular shape with a 10-mm × 15-mm opening at the edge) were surgically attached to the skull under aseptic conditions and general anesthesia. The center of the recording chamber was located 22 mm lateral and 10 mm anterior in stereotaxic coordinates. These coordinates were equivalent to those where gloss-selective neurons were recorded in our previous study, targeting the lower bank of the STS in the mid-IT cortex. The position of the lower bank of the STS was identified using preacquired MRI structural images and depth profiles during electrode insertion. After surgery, the monkey was allowed to recover for at least 1 week before experiments began. During the week following surgery, antibiotics (cefazolin sodium, 12.5 mg/kg) were administered every 12 hours. Recording experiments were conducted in the left hemisphere. Neurons were recorded extracellularly using tungsten microelectrodes (Frederick Haer) inserted vertically with a hydraulic microdrive (MO-951, Narishige) from the vertex through guide tubes fixed to a plastic grid with holes placed at 1-mm intervals. In each electrode penetration, neural responses were examined at multiple depths. By using two grids shifted 0.5 mm vertically and horizontally relative to each other, we achieved a minimum interval of 0.7 mm between holes. During experiments, the monkey performed a fixation task, gazing at a small light spot presented at the screen center. Each trial began with a fixation spot presentation, followed by three stimulus presentations at the visual field center within each trial, with 300-ms interstimulus intervals. Each stimulus presentation lasted 300 ms. After the monkey stably gazed at the fixation spot, it was turned off to avoid interference between the fixation spot and the visual stimulus. If the monkey’s gaze remained within the fixation window (±1° to 1.3° of visual angle) throughout the trial, the monkey received a liquid reward. Eye position was monitored using an infrared eye camera system (ISCAN). Neural responses were analyzed only for correct trials, with five repetitions of each stimulus. All animal care and experimentation procedures followed the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and were approved by the Animal Experimentation Committee of Tamagawa University.Neural data analysisRecorded neural activity was carefully sorted offline and isolated as single units. Mean firing rates were computed for a 300-ms period beginning 50 ms after stimulus onset. We then subtracted baseline activities computed during the period from 300 ms before to 50 ms after the onset of the first stimulus within each trial, and the resultant rate was taken as the measure of neuronal response to visual stimuli. Only neurons showing significant activity increases in response to at least one stimulus (p < 0.05, t-test) were included in the sample of visually responsive neurons.Tests of stimulus selectivityFor each recorded neuron, we first tested responses with the σ t /α stimulus set using three shapes and ### Page 12 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 12determined the optimal shape in which the maximum response was the largest across the three shapes. For a small number of neurons, a 2 × 2 stimulus array (σ t : 0.1 and 1.6; α: 0.01 and 0.3) was used instead of the 3 × 3 array. Then, using stimuli with the optimal shape, neural responses were recorded with both the σ t /α stimulus set and the σ t /α randomized set (σ t /α test), as well as the σ t /albedo stimulus set and σ t /albedo randomized set (σ t /albedo test). The presence or absence of stimulus selectivity was determined through offline quantitative analysis conducted in the following steps. We first conducted two-way ANOVA to determine the presence or absence of main effects for each parameter comprising the σ t /α and σ t /albedo stimulus sets. The effect of each parameter (σ t , α, albedo) was considered significant if the ANOVA test yielded a main effect for that parameter (p < 0.05) in the original image set (σ t /αor σ t /albedo stimulus set), while the shuffled image set (σ t /α or σ t /albedo randomized set) yielded no response or showed no correlation between responses to the shuffled and original image sets evaluated by computing Pearson’s correlation coefficient. The effect of σ t was considered meaningful only if results were consistent between σ t /α and σ t /albedo tests. We then assessed the presence or absence of selectivity for each parameter as follows: Extinction coefficient (σ t ): If the effect of σ t was significant in both σ t /α and σ t /albedo tests, the unit was judged to have selectivity for σ t . Surface roughness (Beckmann α): If the effect of α was significant in the σ t /α test and the results on σ t were consistent between both tests, the unit was judged to have selectivity for α. Albedo (σ s /σ t ): If the effect of albedo was significant in the σ t /albedo test and the results on σ t were consistent between both tests, the unit was judged to have selectivity for albedo.Evaluation of the stimulus preference and the sharpness of the selectivityTo evaluate each neuron’s stimulus preference, maximum response parameters were calculated as follows. Original response data for the nine recorded stimuli were interpolated on σ t –α or σ t –albedo planes by cubic spline interpolation using the interp2 function in MATLAB to 121 points (11 × 11 points on each axis), and the σ t and α or albedo values showing peaks on interpolated maps were calculated and regarded as optimal parameters for each unit. Additionally, interpolated maps were normalized by each unit’s maximum response value, and heatmaps of responses on σ t –α or σ t –albedo planes were created by summing the interpolated maps of neurons included in the considered population. To quantify the sharpness of the stimulus selectivity, we calculated a sparseness index (Rolls & Tovee, 1995; Vinje & Gallant, 2000), which was defined as Sparseness Index =⎡ ⎣1 −(∑n i=1ri n)2(∑n i=1( r2i n)) ⎤ ⎦ / (1 − 1/n) where r i is the firing rate elicited by the ith stimulus in the set of n stimuli. If r i was negative, it was replaced with 0. This index indicates the degree to which responses are unevenly distributed across the set of stimuli. When all stimuli evoke the same response amplitude, the sparseness index is at its minimum, with a value of 0. As the stimulus selectivity sharpens, the index becomes larger and approaches its maximum of 1.ResultsStimulus selectivity of neuronsWe recorded activities from 179 units in the lower bank of the STS in the mid-IT cortex. Of these, we could complete both σ t /α and σ t /albedo tests in 92 units, and 83 of these units exhibited significant responses to at least one of the stimuli in each test. We examined stimulus selectivity in these 83 units. We first present examples of neurons that showed selectivity for some stimulus parameters in these tests. Figure 6 shows responses of an example unit (Unit 1) for which selectivity for σ t , α, and albedo was examined. Figure 6A shows the response to the σ t /αstimulus set using peri-stimulus time histograms (a) and bubble plots (b). The object shape was S2, which was optimal for this unit. Only certain stimuli evoked strong responses, specifically those with small σ t and small αvalues. Considering psychophysical experiment results, stimuli with such parameters are perceived as highly transparent. Two-way ANOVA on these responses showed a significant main effect for σ t (p < 0.001) but not for α (p = 0.13). Figure 6B shows the response to the σ t /albedo stimulus set for the same unit. As with the σ t /α test, strong responses tended to occur when σ t was small. Two-way ANOVA showed significant main effects for both σ t and albedo (σ t : p < 0.001; albedo:p < 0.01). Results of the cubic spline interpolation of these responses are shown in Figure A2 in Appendix B with response contours. In contrast, responses to the randomized sets with shuffled images (C and D) were very weak in both σ t /α and σ t /albedo tests, showing no significant main effects for any parameter or interaction (p > 0.1) and no significant correlation with responses to the original unshuffled image sets (σ t /α: r = 0.22,p > 0.5; σ t /albedo: r = 0.037, p > 0.9). Although average luminance varied with stimuli in both sets, the selective response of this unit cannot be explained by ### Page 13 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 13A C B D a b a b stimulus set t / albedo stimulus set randomized set t / albedo randomized set unit 1 0.3 0.1 0.01 0.1 0.4 1.6 50 spk/s 0.3 0.1 0.01 0.1 0.4 1.6 50 spk/s 0.1 0.5 0.9 0.1 0.4 1.6 0.1 0.4 1.6 0.1 0.5 0.9 100 ms 50 spks/s 100 ms 50 spks/s #162 a b a b extinction coefficient surface roughness extinction coefficient extinction coefficient surface roughness100 ms 50 spks/s 100 ms 50 spks/s extinction coefficient albedo 0 300ms 0 300ms 0 300ms 0 300ms 50 spk/s 50 spk/s albedoFigure 6. Responses of an example neuron (Unit 1) that exhibited selectivity for stimulus parameters. (A) Responses to the σ t –αstimulus set. (a) The responses are depicted as peri-stimulus time histograms (PSTHs). The horizontal bars under the PSTHs indicate the stimulus presentation period. (b) Response of Unit 1 to each stimulus in the σ t –σ stimulus set, depicted as a bubble plot in which response magnitudes are represented by the diameter of circles. Gray indicates a negative response. Arrangement of parameters in a 3 × 3 matrix is the same between a and b. This neuron strongly responded to stimuli with low σ t and low α (sparseness index, 0.449). (B) Responses to the σ t –albedo stimulus set. Other conventions are the same as in A. This neuron strongly responded to stimuli with low σ t and low albedo (sparseness index, 0.535). (C, D) Responses to the σ t –α randomized set (C) and σ t –albedo randomized set (D). Other conventions are the same as in A and B. This neuron responded only weakly to these stimulus sets. This neuron selectively responded to σ t and albedo, but not to α (see Results for details of the analysis of the selectivity). Asterisks indicate the stimulus that evoked the maximum response in a given stimulus set.luminance differences, since no significant main effects were observed with shuffled images. The 3D impression of the object is lost in the shuffled images, and this may have affected the change in responses. However, the selective responses to the original unshuffled image cannot be explained by this because all stimuli in this set have similar impressions of 3D shape. Based on these results, this unit was judged to be selective for σ t and albedo. Stimulus selectivity varied among neurons. Figure 7 shows examples of three units displaying various selectivities. Responses to the σ t /α stimulus set (a),σ t /α randomized set (b), σ t /albedo stimulus set (c), and σ t /albedo randomized set (d) are shown by the bubble plots. Shape S2 was used for all units. In theσ t /α stimulus set, the unit in A (Unit 2) showed a significant main effect only for σ t (σ t : p < 0.001; α: p> 0.6), while units in B and C (Units 3 and 4) showed significant main effects for both σ t and α (Unit 3: p< 0.01 for both σ t and α; Unit 4: p < 0.001 for bothσ t and α). Correlations with responses to the shuffled image set were not significant for all three units (Unit 2: r = −0.04, p = 0.93; Unit 3: r = 0.15, p = 0.71; Unit 4: r = −0.12, p = 0.75). In the σ t /albedo stimulus set, two-way ANOVA showed significant main effects for σ t (p < 0.001) in all units. While Unit 4 showed a significant main effect for albedo (p < 0.001), Units 2 and 3 showed no significant albedo main effects (p> 0.1). Correlations with responses to the shuffled image set were not significant for all three units (Unit 2: r = −0.17, p = 0.65; Unit 3: r = −0.07, p = 0.86; Unit 4: r = 0.26, p = 0.51). Based on these results, we assessed each unit’s selectivity. Unit 2 (Figure 7A) was judged selective only for σ t , but not for α and albedo. Unit 3 (Figure 7B) was judged selective for both σ t and α, but not for albedo. Unit 4 (Figure 7C) was judged selective for all three parameters: σ t , α, and albedo. ### Page 14 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 140.3 0.1 0.01 0.1 0.4 1.6 0.3 0.1 0.01 0.1 0.4 1.6 0.3 0.1 0.01 0.1 0.4 1.6 0.1 0.4 1.6 0.3 0.1 0.01 0.1 0.4 1.6 0.3 0.1 0.01 0.1 0.4 1.6 A B C a b a b a b surface roughness extinction coefficient unit 2 unit 3 unit 4 0.3 0.1 0.01 25 spk/s albedo 0.1 0.5 0.9 0.1 0.4 1.6 0.1 0.4 1.6 0.1 0.5 0.9 #136 c d 0.1 0.5 0.9 0.1 0.4 1.6 0.1 0.4 1.6 0.1 0.5 0.9 #171c d 0.1 0.5 0.9 0.1 0.4 1.6 0.1 0.4 1.6 0.1 0.5 0.9 #176c d 60 spk/s 60 spk/s 60 spk/s 60 spk/s 70 spk/s 70 spk/s 70 spk/s 70 spk/s 25 spk/s 25 spk/s 25 spk/sFigure 7. (A) Responses of another neuron (Unit 2) to the σ t –α stimulus set (a), σ t – α randomized set (b), σ t –albedo stimulus set (c), and σ t –albedo randomized set (d) plotted as bubble plots in the same format as in Figure 6. This neuron selectively responded to σ t , but not to α and albedo (sparseness index, 0.192 for σ t –α set and 0.145 for σ t –albedo set). (B) Responses of another neuron (Unit 3) that selectively responded to σ t and α, but not to albedo (sparseness index, 0.478 for σ t –α set and 0.706 for σ t –albedo set). (C) Responses of another neuron (Unit 4) that selectively responded to σ t , α, and albedo (sparseness index, 0.255 for σ t –α set and 0.274 for σ t –albedo set). Conventions for B and C are the same as in A. Gray indicates a negative response.Figure 8 shows a Venn diagram of the number of units showing selectivity for each parameter among the 83-unit sample. Thirty-nine units (47.0%) showed selectivity for at least one parameter. Of these, 24 units (28.9%) showed selectivity for σ t , 20 units (24.1%) showed selectivity for α, 20 units (24.1%) showed selectivity for albedo, and 6 units (7.2%) showed selectivity for all three parameters. These results demonstrate that neurons in this area exhibit selectivity for various combinations of σ t , α, and albedo.Optimum stimulus and sharpness of selectivityTo visualize how selective units responded to each parameter, we computed and visualized the parameters showing maximum responses in 2D coordinates (Figures 9A, 9B) and the 50% contours of maximum responses (Figures 9C, 9D). Figures 9A, 9C shows results from 32 units exhibiting selectivity for at least one of σ t and α (Figure 8). Maximum response parameters were obtained by cubic spline interpolation (see Methods for details), and coordinatest (n=24) (n=20) albedo (n=20) 9 6 4 3 6 selective unit (n=39) 7 4Figure 8. Number of neurons selective for each of the parameters tested in the recording experiment, plotted as Venn diagrams. Left circle represents neurons selective for σ t (n =24), upper right circle those selective for α (n = 20), and lower right circle those selective for albedo (n = 20). Overlapping parts of circles represent neurons selective for more than one parameter, and overlap of three circles represents neurons selective for all three parameters (n = 6). Number of neurons tested for selectivity was 83. ### Page 15 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 150 0.15 0.3 0.45 0.6 0.75 0.9 0 2 4 6 8 10 120.04 0.88 0 0.5 1 albedo extinction coefficient 1 0.1 albedo 0.05 0.95surface roughness 1 0.1 0.01 extinction coefficient 1 0.1 0.3 0.055 0.01 0.1 0.4 1.6 extinction coefficient surface roughness 0.1 0.5 0.9 0.1 0.4 1.6 extinction coefficient albedo A B C D E F sparseness indexsparseness index number of neurons sparseness index 0 2 4 6 8 10 12 number of neurons 0 0.15 0.3 0.45 0.6 0.75 0.9 sparseness index5 10 15 20 25 305 10 15 20 25 30Figure 9. (A) Distribution of the optimum parameter values where each neuron maximally responded to the test using the σ t –αstimulus set. Vertical axis represents σ t , and horizontal axis represents α. Neurons that exhibited selectivity for at least one of these parameters are depicted (n = 32). Circle represents neurons only selective for σ t , triangle those only selective for α, and square those selective for both σ t and α. Gray value of each symbol indicates the sparseness index of each neuron, as shown in the scale. If more than one neuron has the same optimum parameter values, the positions of symbols are slightly jittered to avoid complete overlap of symbols. (B) Distribution of the optimum parameter values for the σ t –albedo stimulus set. Vertical axis is for σ t , and horizontal axis is for albedo. Neurons that exhibited selectivity for at least one of these parameters are depicted (n = 35). Circle represents neurons only selective for σ t , triangle those only selective for albedo, and square those selective for both σ t and albedo. Other conventions are the same as in A. (C) Superimposed plot of the 50% response contour of the same group of neurons as in A. Vertical axis represents σ t , and horizontal axis represents α. Inside of the response contour of each neuron is shaded in gray. Gray value of each point indicates the number of neurons superimposed, as shown in the scale. The fact that the whole square region has a gray color indicates that the entire parameter range was covered by the response contour of at least one neuron. (D) Superimposed plot of the 50% response contour of the same group of neurons as in B. Vertical axis represents σ t , and horizontal axis represents albedo. Other conventions are the same as in C. (E) Distribution of the sparseness index of the same group of neurons as in A and C (median =0.393). Horizontal axis represents the sparseness index, and vertical axis represents the number of neurons. A larger value of the sparseness index indicates sharper selectivity. (F) Distribution of the sparseness index of the same group of neurons as in B and D(median = 0.345). Other conventions are the same as in E.were plotted on the σ t (vertical axis)–α (horizontal axis) plane (Figure 9A). The horizontal axis uses a logarithmic scale. Stimulus coordinates showing maximum responses tend to distribute along the plane edges, especially on the right side. The plane sides represent locations where σ t or α has the largest or smallest values, and the rightward bias indicates many units showed strong responses to stimuli with small α values—that is, stimuli with strong gloss. Symbol gray values indicate sparseness indices (see scale), with units having sharper selectivity appearing darker. No clear relationship between the sharpness of selectivity and distribution was observed. The diagram in Figure 9C, where 50% response contours of each unit (see Appendix B for Unit 1 depicted in Figure 6 as an example) are superimposed, also shows contour concentration near plane edges. The gray coloring throughout the entire square region indicates that the entire range of the σ t –α plane examined was covered by at least one neuron’s response contour. The histogram in Figure 9E shows the distribution of the sparseness index for these neurons. As a reference for what selectivity corresponds to certain sparseness index values, each unit’s sparseness index is indicated in the ### Page 16 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 16legends of Figures 6 and 7. While selective units display varying degrees of selectivity and sharpness, responses of most units were localized to certain stimulus ranges. Similarly, for 35 units exhibiting selectivity for at least one of σ t and albedo (Figure 8), the coordinates of the maximum response parameters are shown in Figure 9B, and 50% response contours are overlaid in Figure 9D. The σ t values of the stimuli that showed the maximum response took various values for high and low albedo (left and right edge of the diagram), but for intermediate albedo, σ t tended to be concentrated at intermediate values. Distribution along the albedo axis was biased: Excluding three neurons showing maximum responses at intermediate albedo = 0.5, more neurons showed maximum responses in the diagram’s left half (low albedo stimuli), with 22 neurons in the left half and 10 in the right half. In the overlaid 50% response contours for each unit (Figure 9D), many contours curve rightward, also indicating denser representation of low albedo stimuli. Nonetheless, responses from the entire selective unit population covered the entire range of the σ t –albedo plane examined. Figure 9F shows the distribution of the sparseness index for these neurons as a histogram. As a reference, each unit’s sparseness index is indicated in the legends of Figures 6 and 7. Although some units display sharp selectivity, such as Unit 3, most selective units have relatively broad selectivity, and the distribution of the sparseness index is centered at lower values compared to the σ t /α stimulus set shown in Figure 9E. Nonetheless, these neurons selectively responded to specific ranges of parameters tested, and different neurons responded to different ranges of parameters. Results shown in Figures 9C, 9D indicate that the selective neuron population in this region can represent a wide range of σ t , α, and albedo values, although biases exist in preferred parameters.Parameter dependency of the population responses of the selective neuronsIn the psychophysical experiments in this study, we examined how physical parameters related to light transmission, scattering, and reflection affect perceived object transparency. As is well established, we confirmed that σ t is the main physical parameter affecting perceived transparency, but we also showed that perceived transparency can be affected by other parameters, even when σ t remains constant. If similar parameter interactions are observed in neuronal responses, this would support the possibility that neurons in this area are involved in transparency perception. One parameter interaction revealed in psychophysical experiments is that perceived transparency changes significantly depending on albedo when stimulus shape is relatively simple (S2 or S7): Even with constant σ t , perceived transparency tends to decrease for stimuli with high albedo (Experiment 1, “Effects of σ t and albedo on perceived transparency” section, Figure 2 and Figure A1 in Appendix A). Interestingly, responses of neurons with selectivity for at least one of σ t or albedo tend to show maximum responses at small albedo, as shown in Figure 9B. To examine how population responses of these neurons depend on σ t and albedo, we computed a heatmap of responses from 31 neurons tested with S2 or S7 shapes out of 35 units selective for at least one of σ t and albedo (Figure 10A). We first normalized neural responses by each unit’s maximum response and then summed them to generate the heatmap. The population response peaks near the stimulus plane’s left side, where albedo is low, and response attenuates toward the right side (high albedo), even for the same σ t . The response also becomes much weaker as σ t increases downward on the plane. This result corresponds well to the pattern of σ t and albedo effects on perceived object transparency examined in psychophysical experiments. Another parameter affecting perceived transparency was object surface roughness (α) (Experiment 1, “Effect of surface roughness on perceived transparency” section, Figure 4). We observed that even with the same σ t , perceived transparency changes dramatically depending on α, and even for objects with small σ t that transmit light well, perceived transparency is significantly reduced when α is large. To examine whether similar dependency occurs in neuronal responses, we analyzed how responses of 32 neurons showing selectivity for at least one of σ t and α depend on these parameters. We first normalized neural responses by each unit’s maximum response and then summed them to generate a heatmap (Figure 10B). The population response becomes weaker as σ t increases downward on the plane. This trend corresponds to the pattern of σ t effects on perceived transparency in psychophysical experiments. With respect to roughness, however, there are two peaks at the left and right borders, where α is the maximum and minimum, respectively. Such dual peaks appear not consistent with the results of psychophysical experiments, where perceived transparency is high when roughness is small. We found that if we restrict the sample of population analysis to the 12 neurons showing selectivity for bothσ t and α among units tested with the σ t –α stimulus set, the population response peaks at the stimulus plane’s upper right with small σ t and small α, and responses are weak to stimuli with the same σ t but largeα (Figure 10C). This trend corresponds well to the pattern of σ t and α effects on perceived transparency investigated in psychophysical experiments. We also found that population responses of neurons selective for at least one of σ t or α also exhibit a pattern that is consistent with the results of psychophysical experiments if the population responses are computed under some feasible assumption (Supplementary ### Page 17 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 17surface roughness 0.1 0.4 1.6 0.3 0.055 0.01B0.1 0.4 1.6 0.1 0.5 0.9 albedo t / albedo testAextinction coefficientextinction coefficient 0.1 0.4 1.6 0.3 0.055 0.01C0.6 0.5 0.4 0.3 0.20.6 0.5 0.4 0.3 0.20.6 0.5 0.4 0.3 0.2 extinction coefficientextinction coefficient extinction coefficientextinction coefficient surface roughnessFigure 10. (A) Response contour of the population of neurons (n = 31) that selectively responded to at least one of σ t and albedo in the test using the σ t –albedo stimulus set with relatively simple shapes (S2 or S7). Vertical axis represents σ t , and horizontal axis represents albedo. (B) Response contour of the population of neurons (n = 32) that selectively responded to at least one of σ t and αin the test using the σ t –α stimulus set. Vertical axis represents σ t , and horizontal axis represents α. (C) Response contour of the population of neurons (n = 12) that selectively responded to both σ t and α in the test using the σ t –α stimulus set. Vertical axis represents σ t , and horizontal axis represents α. In each panel, response contours were computed as the sum of the normalized response of each neuron. Contour colors represent response magnitudes normalized by the number of neurons. Yellow represents a strong population response, and blue indicates a weak response.0.23 >= σt 0.69 >= σt > 0.23σt > 0.69 0.027 >= α0.108 >= α > 0.027α > 0.108 albedo >= 0.66 0.66 > albedo >= 0.34 0.34 > albedo A B Cextinction coefficient surface roughness albedo M L L20 L25 M L L20 L25 M L L20 L25 A15 A10 A5 A P A15 A10 A5 A P A15 A10 A5 A P fundus of STS lip of STSFigure 11. Recording sites of the selective neurons overlaid with the top view of the lower bank of the STS in the left hemisphere (copied from a previous study; Nishio et al. 2012) with stereotaxic coordinates. A is for σ t , B is for α, and C is for albedo, respectively. Open circles indicate the positions of the grid holes for electrode penetrations, and a colored dot indicates a selective neuron recorded at a given coordinate. Different colors of the dots represent a different range of the optimum parameter of each neuron. The number of color symbols in each panel corresponds to the number of neurons that exhibited selectivity for each parameter that is shown in the Venn diagram in Figure 8 (24 for σ t , 20 for α, 20 for albedo). If multiple color symbols are plotted in a single circle, each neuron was recorded at a different depth. In A, the parameter value in the test using the σ t –α stimulus set is used. Gray contour indicates the recording area at which gloss-selective neurons were searched in a previous study (Baba et al., 2021).Material 8). Thus, we found that surface roughness effects on perceived transparency also correspond to responses from neuronal populations selective for bothσ t and α or, with some assumptions, to activity from larger neuronal populations selective for at least one of these parameters.Distribution of selective neuronsIn Figure 11, recording positions of neurons exhibiting selectivity for σ t , α, and albedo are shown in panels A, B, and C, respectively. Circles correspond to grid hole positions (see Methods), and dots indicate recording positions of neurons showing selectivity ### Page 18 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 18for each parameter. They are located on the lower bank of STS in the middle portion of the IT cortex. The area connected by the solid line shows the region where gloss-selective neurons were examined in a previous study (Baba et al., 2021) as a reference. In this experiment, neurons were sampled from the medial part of this area and the adjacent posterior region. Neurons selectively responding to each parameter were distributed across the entire examined area, but many were recorded from the recording area’s posterior part. No clear distributional differences existed between parameters, suggesting that units selective for these three parameters are intermixed in this area. Dot colors represent the range of each parameter divided into three levels and indicate each unit’s optimal parameter value. No clear distributional differences were observed for maximum response values of each parameter, as indicated by dot colors. However, larger neuronal samples from multiple hemispheres would be necessary to obtain more precise perspectives on these issues.DiscussionNeuronal selectivity for parameters related to transparency/translucencyIn this study, we found neurons in the lower bank of STS that selectively respond to extinction coefficient (σ t ), which is strongly associated with perceived object transparency/translucency. We also found neurons in the same area that selectively respond to other parameters shown to affect perceived transparency/translucency in psychophysical experiments, namely, surface roughness (α) and albedo. Each neuron selectively responded to a specific range of the parameter space, and as a population, the responses of selective neurons covered the entire range of the parameter space examined. The recording site was near the area where we previously recorded gloss-selective neurons (Baba et al., 2021; Nishio et al., 2012). Both gloss and transparency are perceptual attributes related to an object’s optical properties and greatly affect object appearance. Therefore, we chose this location as an initial attempt to search for neurons expressing transparency/translucency-related parameters. We consider that α-selective neurons in this experiment likely have similar properties to previously recorded gloss-selective neurons, since α is a gloss-related parameter and was recorded near the area where many gloss-selective neurons were previously found, although we cannot specify exact correspondences because selectivity examination methods differ. Additionally, albedo is a parameter related to object lightness, and Nishio, Shimokawa, Goda, and Komatsu (2014) showed that gloss-selective neuron populations carry information about object lightness as well as highlight intensity and sharpness. The distribution of neurons showing selectivity in this experiment (Figure 11) overlaps with the area where gloss-selective neurons were recorded in previous studies and extends slightly to more posterior regions. Previous studies also showed individual variation in gloss-selective neuron distribution, and the posterior extension may be due to individual differences. The distribution of neurons selective for σ t , α, and albedo overlapped, and many neurons showed selectivity for multiple parameters (Figure 8). These results suggest that many neurons carrying glossiness and transparency-related information are concentrated in this area. Chadwick, Heywood, Smithson, and Kentridge (2019) reported that a patient with severe color vision and texture perception impairment due to brain damage could still distinguish liquid tea strength and milk concentration to some extent, suggesting that the perceptual ability of transparency may be preserved. This patient also had preserved gloss perception (Kentridge, Thomson, & Heywood, 2012), and our results showing that gloss and transparency-related parameters are represented in the same cortical area may explain their observations in brain-damaged human patients. We should note that we do not think this area alone represents transparency/translucency-related information. Regarding gloss, an fMRI study showed activity in the middle and posterior portions of the IT cortex (Okazawa et al., 2012), and neuroanatomical studies (Komatsu, Nishio, Ichinohe, & Goda, 2021) and neuron activity manipulation experiments (Baba et al., 2021) have suggested that multiple IT cortex subareas represent gloss information and may form a network. Similar observations exist for other important visual attributes such as faces (Arcaro, Schade, & Livingstone, 2019; Moeller, Freiwald, & Tsao, 2008; Tsao, Moeller, & Freiwald, 2008) and colors (Banno, Ichinohe, Rockland, & Komatsu, 2011; Conway, Moeller, & Tsao, 2007; Harada et al., 2009; Lafer-Sousa & Conway, 2013; Lafer-Sousa, Conway, & Kanwisher, 2016), and we speculate that a network of multiple areas involved in neural processing of transparency/translucency also exists and may relate to representing different aspects of transparency/translucency.Correlation between neuronal activity and perceptionIn this study, we tested neuronal selectivity for stimulus sets combining σ t and α, as well as stimulus sets combining σ t and albedo, allowing us to investigate parameter interactions in neuronal responses and compare them with parameter interactions in psychophysical experiments. One interaction revealed in psychophysical experiments is that for objects with relatively simple shapes, high albedo reduces perceived transparency even when σ t is constant. In neural recording experiments, we found that population ### Page 19 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 19responses of neurons selective for either σ t or albedo peak at low albedo and attenuate as albedo increases, even for the same σ t (Figure 10A). This pattern is similar to the albedo effects on perceived transparency found in psychophysical experiments. The same heatmap shows that responses to stimuli with large σ t were very weak, consistent with the idea that these neuronal responses relate to transparency perception, but the maximum response occurred around σ t 0.3 rather than at the smallestσ t . Presumably, this is due to summing responses from neurons with various properties, as shown in Figure 7. Additionally, since translucent objects with some absorption and scattering are more common than highly transparent stimuli, neuronal population response properties may be optimized for analyzing more general transparent/translucent objects. Another parameter interaction examined in neuronal population responses was that of α and σ t . In psychophysical experiments, objects with large αsignificantly reduced perceived transparency even whenσ t was constant. We found that population responses of neurons selective for both σ t and α peaked at lowα and attenuated as α increased, even for the same σ t (Figure 10C). This pattern is similar to the effect ofα on perceived transparency found in psychophysical experiments. Maximum response occurred aroundσ t 0.3 rather than at the smallest σ t , similar to the albedo case. We also found that similar results could be obtained for larger neuronal populations showing selectivity for at least one of σ t and α by summing responses under appropriate assumptions (Supplementary Material 8). We believe that the fact that interaction patterns between σ t and albedo, as well as between σ t and α observed in population responses of neurons selective for these parameter sets, are similar to those observed in psychophysical experiments supports the idea that responses from neurons selective for these parameters in this area are related to transparency/translucency perception.General discussionIn this study, we psychophysically examined the properties of transparency/translucency perception in human participants using stimuli that manipulated the extinction coefficient, surface roughness, albedo, and phase function—physical parameters related to the transmission, reflection, and scattering of light on the surface and within an object. We also examined the responses of neurons in the macaque IT cortex to these stimuli. While it is generally thought that the visual system processes image features that carry information about physical parameters, it is highly unlikely that the physical parameters listed above are directly processed in the brain. However, it remains unclear which specific image features are related to transparency/translucency perception. Given the early stage of this research, we believe it is an effective strategy to use stimuli based on well-defined physical parameters, allowing us to explore a well-defined target space systematically. Transparency and translucency of objects carry important information about the states of natural objects, such as food safety or the juiciness of fruit (Di Cicco, Wijntjes, & Pont, 2020; Ray & Roberts, 2013). It has been argued that there may exist a specialized region of the cerebral cortex for processing translucence (Chadwick et al., 2019, p. 211). Considering that transparency/translucency is a perceptual attribute useful for judging an object’s state, it is reasonable that these attributes are processed in the ventral pathway involved in object vision. Similarly, glossiness of objects plays an important role in judging an object’s state. Thus, our finding that information about glossiness and transparency/translucency is processed in the same area of the ventral higher visual cortex seems reasonable. However, the visual information related to glossiness and transparency/translucency differs in many ways, and there should be differences in the neural processing of these two attributes. As described below, various cues are involved in transparency/translucency processing, and divergent brain areas may be involved.Visual cues for transparency/translucency perceptionTransparent objects create a distorted image of the background due to the pattern of light refracted and transmitted through the object. The distortion of the background image compared to the surrounding field provides clues to perceived transparency and creates a sense of glass-like materiality (Fleming, Jakel, & Maloney, 2011). Gkioulekas et al. (2015) showed that translucent objects exhibit characteristic patterns of luminance change at edges where surface orientation changes discontinuously, suggesting the importance of edge information in translucency perception. Tamura, Higashi, and Nakauchi (2018) demonstrated that information about the internal motion field of a rotating object is used to distinguish between glass and mirror-like objects. Caustics are also effective cues for perceived transparency/translucency, even when the cue is located remotely from an object (Gigilashvili, Dubouchet, et al., 2020). The stimuli used in our experiments were static images of smooth objects without backgrounds and did not contain any of the above cues. Therefore, the results of this study cannot provide direct insights into how these cues are processed in the brain. However, we can make some inferences based on other studies. ### Page 20 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 20Neural processing of different transparency cuesThe distorted background image created by light refracted and transmitted through an object generates texture patterns with varying statistics. It has been suggested that texture processing is performed in the ventral visual pathway (Arcizet, Jouffrais, & Girard, 2008; Freeman, Ziemba, Heeger, Simoncelli, & Movshon, 2013; Kim, Bair, & Pasupathy, 2022; Köteles, De Maziere, Van Hulle, Orban, & Vogels, 2008; Okazawa, Tajima, & Komatsu, 2015) and is separate from gloss processing (Kentridge et al., 2012). It is possible that transparency caused by background distortion is related to texture processing, which may be performed in a different region of the ventral visual pathway from the one we examined in this study. The edge features of translucent objects identified by Gkioulekas et al. (2015) described in the previous section are relatively simple stimulus features and may be detected in early visual cortex. Qiu and von der Heydt (2007) reported that border ownership–sensitive cells in macaque area V2 showed responses that depended on the transparency of overlying objects. Although they used simple stimuli in which uniform light and dark squares were combined to create the appearance of transparency rather than realistic object images that reproduced light behavior, their observations indicate that the classification of edges with respect to the transparency of overlapping objects already begins at the V2 stage. The edge features of translucent objects identified by Gkioulekas et al. may also be represented at this stage. Perception of transparency/translucency through caustics projected on separate surfaces is achieved by integrating relationships between separate elements in an image, such as the object and the light and dark patterns in shadows. How such separate elements are integrated to create the appearance of an object is an interesting question and may involve more complex processing than other cues. Perception of transparency related to motion signals likely involves areas such as middle temporal area (MT) and floor of the superior temporal sulcus area (FST), where neurons sensitive to motion transparency have been reported (Qian & Andersen, 1994; Rosenberg, Wallisch, & Bradley, 2008; Stoner & Albright, 1992). These areas may also be involved when discriminating between glass and mirror-like objects using information about the motion field inside a rotating object (Tamura et al., 2018). We used static images of relatively thick objects without sharp edges, and it is unlikely that those cues mentioned above play important roles in the transparency/translucency of our stimuli. Presumably, the relationship between the contrast and blur of specular highlights and nonspecular regions as suggested by Motoyoshi (2010) may play important roles. It would be interesting to examine how the manipulation of such cues affects neural responses in this area in the future study.Insights from studies using deep learningDeep neural networks (DNNs), trained to recognize object images, have been shown to become good models of neural processing for object recognition in the ventral visual areas of the human and nonhuman primate brain (Eickenberg, Gramfort, Varoquaux, & Thirion, 2017; Lindsay, 2020; Storrs & Kriegeskorte, 2021; Yamins et al., 2014). Many early studies in this field used supervised DNN models. Although the functional architecture of deep convolutional networks often used in these models resembles the brain’s processing, it differs in the need for large amounts of image labeling. On the other hand, unsupervised DNNs, which train without labeling, have also demonstrated high performance in object recognition (Zhuang et al., 2021). Higgins et al. (2021) showed that neuronal responses in face patches in monkey IT cortex closely correspond to the latent variables of a model trained on face images using a beta-variational autoencoder (β-VAE) (Higgins et al., 2017), an unsupervised learning model. In material perception, object images are formed by the complex interactions among optical properties, illumination, and the shape of an object, and how these factors are separated is an important question. Storrs, Anderson, and Fleming (2021) trained an unsupervised DNN called PixelVAE (Gulrajani et al., 2016) using images of high- and low-gloss bumpy surfaces under various lighting conditions and found that the VAE’s latent space separately represented information such as glossiness, illumination, and shape, even without providing any labels for these factors. They also showed that the image features closely related to human gloss perception can be decoded from the network’s latent space. More recently, Liao, Sawayama, and Xiao (2023) trained an unsupervised model combining a style-based generative adversarial network (StyleGAN) (Karras, Laine, & Aila, 2021) and the pixel2style2pixel (pSp) encoder (Richardson et al., 2020) on a large number of translucent soap images and found that color, translucency, and shape information were separately represented as latent variables in different layers of the model. These layers also had different spatial scales, with the layer representing translucency having an intermediate spatial scale. While it is still difficult to relate the mechanisms of the models used in these studies to the brain, it is an interesting question how the response properties of IT neurons sensitive to glossiness and/or transparency/translucency we observed are related to the latent variables of these models. Furthermore, by manipulating latent variables ### Page 21 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 21associated with specific factors in the generative models used in these studies, one can systematically manipulate specific attributes related to material perception, such as translucency. Image sets generated in this way may be very useful as stimuli in studies examining brain responses (Liao et al., 2023).Conclusions and future directionsAs described above, various cues are involved in transparency/translucency processing. The present study demonstrated the possibility that transparency/translucency processing may occur in the higher ventral visual cortex using static cues present within object images. Considering that transparency/translucency perception is useful for judging an object’s material properties and state, neurons in this area may function to utilize transparency/translucency information in object vision. Other brain regions are thought to be involved in processing the various transparency/translucency cues mentioned above. An important theme for future research will be to clarify which brain regions are involved in processing those cues and how they relate to the processing in the IT cortex region described in the present study.Keywords: transparency, translucency, perception, IT cortex, macaque monkey, neuronal selectivityAcknowledgmentsThe authors thank T. Ota for technical assistance with monkey treatment, A. Masaoka for technical assistance in setting the recording system, D. Kato for providing code for spike sorting, T. Shimokawa for providing 3D shapes, A. Nishio for assistance with the generation of transparent object images, and E. Watanabe for help in checking English using Claude Sonnet 4. Supported by JSPS KAKENHI grant numbers 15H05916 (Grant-in-Aid for Scientific Research on Innovative Areas “Innovative SHITSUKAN Science and Technology”), 20H05955 (Grant-in-Aid for Transformative Research Areas (A) “Deep Shitsukan”) and 26K15033 (Grant-in-Aid for Scientific Research (C)). Commercial relationships: none. Corresponding author: Hidehiko Komatsu. Email: hkomatsu@lab.tamagawa.ac.jp. Address: Brain Science Institute, Tamagawa University, Machida, Tokyo 194-8610, Japan.ReferencesArcaro, M. J., Schade, P. F., & Livingstone, M. S. (2019). Universal mechanisms and the development of the face network: What you see is what you get.Annual Review of Vision Science, 5, 341–372, https: //doi.org/10.1146/annurev-vision-091718-014917. Arcizet, F., Jouffrais, C., & Girard, P. (2008). Natural textures classification in area V4 of the macaque monkey. Experimental Brain Research, 189(1), 109– 120, https://doi.org/10.1007/s00221-008-1406-9. Baba, M., Nishio, A., & Komatsu, H. (2021). Relationship between the activities of gloss-selective neurons in the macaque inferior temporal cortex and the gloss discrimination behavior of themonkey. Cerebral Cortex Communications, 2(1), 1–13, https://doi.org/10.1093/texcom/tgab011. Banno, T., Ichinohe, N., Rockland, K. S., & Komatsu, H. (2011). Reciprocal connectivity of identified color-processing modules in the monkey inferior temporal cortex. Cerebral Cortex, 21(6), 1295–1310, https://doi.org/10.1093/cercor/bhq211. Beck, J., & Ivry, R. (1988). On the role of figural organization in perceptual transparency. Attention, Perception, & Psychophysics, 44(6), 585–594, https://doi.org/10.3758/bf03207492. Brainard, D. H. (1997). The Psychophysics Toolbox.Spatial Vision, 10(4), 433–436. Chadwick, A. C., Heywood, C. A., Smithson, H. E., & Kentridge, R. W. (2019). Translucence perception is not dependent on cortical areas critical for processing colour or texture.Neuropsychologia, 128, 209–214, https: //doi.org/10.1016/j.neuropsychologia.2017.11.009. Chowdhury, N. S., Marlow, P. J., & Kim, J. (2017). Translucency and the perception of shape. Journal of Vision, 17(3), 17, https://doi.org/10.1167/ 17.3.17. Conway, B. R., Moeller, S., & Tsao, D. Y. (2007). Specialized color modules in macaque extrastriate cortex. Neuron, 56(3), 560–573, https://doi.org/10.1016/j.neuron.2007.10.008. Di Cicco, F., Wijntjes, M. W. A., & Pont, S. C. (2020). If painters give you lemons, squeeze the knowledge out of them. A study on the visual perception of the translucent and juicy appearance of citrus fruits in paintings. Journal of Vision, 20(13), 12, https://doi.org/10.1167/jov.20.13.12. Eickenberg, M., Gramfort, A., Varoquaux, G., & Thirion, B. (2017). Seeing it all: Convolutional network layers map the function of the human visual system. Neuroimage, 152, 184–194, https://doi.org/10.1016/j.neuroimage.2016.10. 001. ### Page 22 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 22Fleming, R. W., & Bülthoff, H. H. (2005). Low-level image cues in the perception of translucent materials. ACM Transactions on Applied Perception, 2(3), 346–382. Fleming, R. W., Jensen, H. W., & Bülthoff, H. H. (2004). Perceiving translucent materials. In Proceedings of APGV 2004: ACM SIGGRAPH Symposium on Applied Perception in Graphics and Visualization(pp. 127–134). New York, NY: ACM Press, https://doi.org/10.1145/1012551.1012575. Fleming, R. W., Jakel, F., & Maloney, L. T. (2011). Visual perception of thick transparent materials. Psychological Science, 22(6), 812–820, https://doi.org/10.1177/0956797611408734. Freeman, J., Ziemba, C. M., Heeger, D. J., Simoncelli, E. P., & Movshon, J. A. (2013). A functional and perceptual signature of the second visual area in primates. Nature Neuroscience, 16(7), 974–981. https://doi.org/10.1038/nn.3402. Gerardin, M., Simonot, L., Farrugia, L. P., Iehl, J. C., Fournel, T., & Hebert, M. (2019). A translucency classification for computer graphics. In Proceedings of Material Appearance 2019, IS&T International Symposium on Electronic Imaging (pp. 203:1–203:6). Burlingame, CA: Society for Imaging Science and Technology, https://doi.org/10.2352/ISSN. 2470-1173.2019.6.MAAP-203. Gigilashvili, D., Dubouchet, L., Pedersen, M., & Hardeberg, J. Y. (2020). Caustics and translucency perception. In Proceedings of Material Appearance 2020, IS & T International Symposium on Electronic Imaging (pp. 033:1–033:6). Burlingame, CA: Society for Imaging Science and Technology, https://doi. org/10.2352/ISSN.2470-1173.2020.5.MAAP-033. Gigilashvili, D., Thomas, J. B., Hardeberg, J. Y., & Pedersen, M. (2020). On the nature of perceptual translucency. In Proceedings of the 8th Annual Workshop on Material Appearance Modeling (MAM2020) (pp. 17–20). Eurographics Digital Library. The Eurographics Association, https://doi.org/10.2312/mam.20201141. Gigilashvili, D., Shi, W. Q., Wang, Z. Y., Pedersen, M., Hardeberg, J. Y., & Rushmeier, H. (2021). The role of subsurface scattering in glossiness perception.ACM Transactions on Applied Perception, 18(3):10, 1–26, https://doi.org/10.1145/3458438. Gigilashvili, D., Thomas, J. B., Hardeberg, J. Y., & Pedersen, M. (2021). Translucency perception: A review. Journal of Vision, 21(8), 4, https://doi.org/10.1167/jov.21.8.4. Gkioulekas, I., Xiao, B., Zhao, S., Adelson, E., Zickler, T., & Bala, K. (2013). Understanding the role of phase function in translucent appearance.ACM Transactions on Graphics, 32(5), 1–19, https://doi.org/10.1145/2516971.2516972. Gkioulekas, I., Walter, B., Adelson, E. H., Bala, K., & Zickler, T. (2015). On the appearance of translucent edges. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (pp. 5528–5536). Boston, MA: IEEE, https://doi.org/10.1109/CVPR.2015.7299192. Gulrajani, I., Kumar, K., Ahmed, F., Taiga, A. A., Visin, F., Vazquez, D., . . . Courville, A. (2016). PixelVAE: A latent variable model for natural images. Preprint at arXiv. https://arxiv.org/abs/1611.05013. Harada, T., Goda, N., Ogawa, T., Ito, M., Toyoda, H., Sadato, N., . . . Komatsu, H. (2009). Distribution of colour-selective activity in the monkey inferior temporal cortex revealed by functional magnetic resonance imaging. European Journal of Neuroscience, 30(10), 1960–1970, https://doi.org/10.1111/j.1460-9568.2009.069 95.x. Higgins, I., Matthey, L., Pal, A., Burgess, C., Glorot, X., Botvinick, M., . . . Lerchner, A. (2017).β-VAE: learning basic visual concepts with a constrained variational framework. In Proceedings of the 5th International Conference on Learning Representations, ICLR (ICLR, 2017) (pp. 1–13). https://openreview.net/forum?id=Sy2fzU9gl. Higgins, I., Chang, L., Langston, V., Hassabis, D., Summerfield, C., Tsao, D., . . . Botvinick, M. (2021). Unsupervised deep learning identifies semantic disentanglement in single inferotemporal face patch neurons. Nature Communications, 12(1), 6456, https://doi.org/10.1038/s41467-021-26751-5. Jacob, W. (2010). Mitsuba renderer. http://www. mitsuba-renderer.org. Karras, T., Laine, S., & Aila, T. A. (2021). Style-based generator architecture for generative adversarial networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(12), 4217–4228, https://doi.org/10.1109/TPAMI.2020.2970919. Kentridge, R. W., Thomson, R., & Heywood, C. A. (2012). Glossiness perception can be mediated independently of cortical processing of colour or texture. Cortex, 48(9), 1244–1246, https://doi.org/10.1016/j.cortex.2012.01.011. Khang, B. G., & Zaidi, Q. (2002). Accuracy of color scission for spectral transparencies. Journal of Vision, 2(6), 451–466, https://doi.org/10.1167/2.6.3. Kim, T., Bair, W., & Pasupathy, A. (2022). Perceptual texture dimensions modulate neuronal response dynamics in visual cortical area V4.Journal of Neuroscience, 42(4), 631–642, https: //doi.org/10.1523/JNEUROSCI.0971-21.2021. Komatsu, H., Nishio, A., Ichinohe, N., & Goda, N. (2021). Structure and function of neural circuit related to gloss perception in the macaque inferior temporal cortex: A case report. Brain ### Page 23 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 23Structure and Function, 226(9), 3023–3030, https://doi.org/10.1007/s00429-021-02324-6. Köteles, K., De Maziere, P. A., Van Hulle, M., Orban, G. A., & Vogels, R. (2008). Coding of images of materials by macaque inferior temporal cortical neurons. European Journal of Neuroscience, 27(2), 466–482, https://doi.org/10.1111/j.1460-9568.2007.06008.x. Lafer-Sousa, R., & Conway, B. R. (2013). Parallel, multi-stage processing of colors, faces and shapes in macaque inferior temporal cortex.Nature Neuroscience, 16(12), 1870–1878, https://doi.org/10.1038/nn.3555. Lafer-Sousa, R., Conway, B. R., & Kanwisher, N. G. (2016). Color-biased regions of the ventral visual pathway lie between face- and place-selective regions in humans, as in macaques.Journal of Neuroscience, 36(5), 1682–1697, https: //doi.org/10.1523/JNEUROSCI.3164-15.2016. Liao, C., Sawayama, M., & Xiao, B. (2023). Unsupervised learning reveals interpretable latent representations for translucency perception.PLOS Computational Biology, 19(2), e1010878, https://doi.org/10.1371/journal.pcbi.1010878. Lindsay, G. W. (2020). Convolutional neural networks as a model of the visual system: past, present, and future. Journal of Cognitive Neuroscience, 33(10), 2017–2031, https://doi.org/10.1162/jocn_a_0 1544. Marlow, P. J., & Anderson, B. L. (2021). The cospecification of the shape and material properties of light permeable materials. Proceedings of the National Academy of Sciences of the United States of America, 118(14), 1–10. https://doi.org/10.1073/pnas.2024798118. Marlow, P. J., Kim, J., & Anderson, B. L. (2017). Perception and misperception of surface opacity.Proceedings of the National Academy of Sciences of the United States of America, 114(52), 13840–13845, https://doi.org/10.1073/pnas.1711416115. Metelli, F. (1974). The perception of transparency.Scientific American, 230(4), 90–98, https: //doi.org/10.1038/scientificamerican0474-90. Moeller, S., Freiwald, W. A., & Tsao, D. Y. (2008). Patches with links: A unified system for processing faces in the macaque temporal lobe. Science, 320(5881), 1355–1359, https: //doi.org/10.1126/science.1157436. Motoyoshi, I. (2010). Highlight-shading relationship as a cue for the perception of translucent and transparent materials. Journal of Vision, 10(9), 6, https://doi.org/10.1167/10.9.6. Nagai, T., Ono, Y., Tani, Y., Koida, K., Kitazaki, M., & Nakauchi, S. (2013). Image regions contributing to perceptual translucency: A psychophysical reverse-correlation study. i-Perception, 4(6), 407–428, https://doi.org/10.1068/i0576. Nishio, A., Goda, N., & Komatsu, H. (2012). Neural selectivity and representation of gloss in the monkey inferior temporal cortex. Journal of Neuroscience, 32(31), 10780–10793, https: //doi.org/10.1523/JNEUROSCI.1095-12.2012. Nishio, A., Shimokawa, T., Goda, N., & Komatsu, H. (2014). Perceptual gloss parameters are encoded by population responses in the monkey inferior temporal cortex. Journal of Neuroscience, 34(33), 11143–11151, https: //doi.org/10.1523/JNEUROSCI.1451-14.2014. Okazawa, G., Goda, N., & Komatsu, H. (2012). Selective responses to specular surfaces in the macaque visual cortex revealed by fMRI. Neuroimage, 63(3), 1321–1333, https: //doi.org/10.1016/j.neuroimage.2012.07.052. Okazawa, G., Tajima, S., & Komatsu, H. (2015). Image statistics underlying natural texture selectivity of neurons in macaque V4. Proceedings of the National Academy of Sciences of the United States of America, 112(4), E351–E360, https://doi.org/10.1073/pnas.1415146112. Pelli, D. G. (1997) The VideoToolbox software for visual psychophysics: transforming numbers into videos. Spatial Vision, 10(4), 437–442. Qian, N., & Andersen, R. A. (1994). Transparent motion perception as detection of unbalanced motion signals. II. Physiology. Journal of Neuroscience, 14(12), 7367–7380, https: //doi.org/10.1523/JNEUROSCI.14-12-07367.1994. Qiu, F. T., & von der Heydt, R. (2007). Neural representation of transparent overlay. Nature Neuroscience, 10(3), 283–284, https://doi.org/10. 1038/nn1853. Ray, A., & Roberts, J. R. (2013). The structural basis of egg shell translucency and its role in food safety of table eggs. In Proceedings of the 24th Annual Australian Poultry Science Symposium (p. 162). Sydney, Australia. Reinhard, E., Stark, M., Shirley, P., & Ferwerda, J. (2002). Photographic tone reproduction for digital images. ACM Transactions on Graphics, 21(3), 267–276, https://doi.org/10.1145/566654.566575. Richardson, E., Alaluf, Y., Patashnik, O., Nitzan, Y., Azar, Y., Shapiro, S., . . . Cohen-Or, D. (2020). Encoding in style: a StyleGAN encoder for image-to-image translation. Preprint at arXiv. https://arxiv.org/abs/2008.00951. Rolls, E. T., & Tovee, M. J. (1995). Sparseness of the neuronal representation of stimuli in the primate temporal visual cortex. Journal ### Page 24 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 24of Neurophysiology, 73(2), 713–726, https: //doi.org/10.1152/jn.1995.73.2.713. Rosenberg, A., Wallisch, P., & Bradley, D. C. (2008). Responses to direction and transparent motion stimuli in area FST of the macaque.Visual Neuroscience, 25(2), 187–195, https: //doi.org/10.1017/S0952523808080528. Sawayama, M., Dobashi, Y., Okabe, M., Hosokawa, K., Koumura, T., Saarela, T. P., . . . Nishida, S. (2022). Visual discrimination of optical material properties: A large-scale study. Journal of Vision, 22(2), 17, https://doi.org/10.1167/jov.22.2.17. Shimokawa, T., Nishio, A., Sato, M. A., Kawato, M., & Komatsu, H. (2019). Computational model for human 3D shape perception from a single specular image. Frontiers in Computational Neuroscience, 13, 10, https://doi.org/10.3389/fncom.2019.00010. Singh, M., & Anderson, B. L. (2002a). Perceptual assignment of opacity to translucent surfaces: The role of image blur. Perception, 31(5), 531–552, https://doi.org/10.1068/p3363. Singh, M., & Anderson, B. L. (2002b). Toward a perceptual theory of transparency. Psychological Review, 109(3), 492–519, https://doi.org/10.1037/ 0033-295X.109.3.492. Stoner, G. R., & Albright, T. D. (1992). Neural correlates of perceptual motion coherence. Nature, 358(6385), 412–414, https://doi.org/10.1038/358412a0. Storrs, K. R., Anderson, B. L., & Fleming, R. W. (2021). Unsupervised learning predicts human perception and misperception of gloss.Nature Human Behaviour, 5(10), 1402–1417, https://doi.org/10.1038/s41562-021-01097-6. Sun, H. C., Ban, H., Di Luca, M., & Welchman, A. E. (2015). fMRI evidence for areas that process surface gloss in the human visual cortex. Vision Research, 109, 149–157, https://doi.org/10.1016/j.visres.2014.11.012. Tamura, H., Higashi, H., & Nakauchi, S. (2018). Dynamic visual cues for differentiating mirror and glass. Scientific Reports, 8(1), 1–12, 8403, https://doi.org/10.1038/s41598-018-26720-x. Tsao, D. Y., Moeller, S., & Freiwald, W. A. (2008). Comparing face patch systems in macaques and humans. Proceedings of the National Academy of Sciences of the United States of America, 105(49), 19514–19519, https://doi.org/10.1073/pnas.0809662105. Vinje, W. E., & Gallant, J. L. (2000). Sparse coding and decorrelation in primary visual cortex during natural vision. Science, 287(5456), 1273–1276, https://doi.org/10.1126/science.287.5456.1273. Wada, A., Sakano, Y., & Ando, H. (2014). Human cortical areas involved in perception of surface glossiness. Neuroimage, 98, 243–257, https://doi.org/10.1016/j.neuroimage.2014.05.001. Xiao, B., Walter, B., Gkioulekas, I., Zickler, T., Adelson, E., & Bala, K. (2014). Looking against the light: how perception of translucency depends on lighting direction. Journal of Vision, 14(3):17, 1–22, https://doi.org/10.1167/14.3.17. Yamins, D. L. K., Hong, H., Cadieu, C. F., Solomon, E. A., Seibert, D., & DiCarlo, J. J. (2014). Performance-optimized hierarchical models predict neural responses in higher visual cortex.Proceedings of the National Academy of Sciences of the United States of America, 111(23), 8619–8624, https://doi.org/10.1073/pnas.1403112111. Zhuang, C., Yan, S., Nayebi, A., Schrimpf, M., Frank, M. C., DiCarlo, J. J., . . . Yamins, D. L. K. (2021). Unsupervised neural network models of the ventral visual stream. Proceedings of the National Academy of Sciences of the United States of America, 118(3), e2014196118, https://doi.org/10.1073/pnas.2014196118.Appendix A. Effects of σt and albedo on perceived transparencyResults of the experiments using the stimulus set depicted in Figure 1 (for shape S7) and Supplementary Figure S1 (for shapes S2 and S11) to examine the effects of extinction coefficient (σ t ) and albedo (σ s /σ t ) on the perceived transparency of objects are shown in Figure 2 in 2D matrix format. Figure A1 shows the same results in graphical form, depicting the relationship between extinction coefficient (σ t ) and the perceived transparency (A), as well as the relationship between albedo (σ s /σ t ) and perceived transparency (B), respectively. In both A and B, average ratings of perceived transparency (0–99) for nine participants (vertical axis) are plotted against σ t or albedo. In A, lines with different colors correspond to different albedo, and in B, lines with different colors correspond to different σ t . Error bars indicate SD.B. Evaluation of the stimulus selectivity of a neuron using cubic spline interpolationResponses of a neuron (Unit 1) whose responses are depicted in Figure 6 are interpolated by cubic spline interpolation using the interp2 function in MATLAB to 121 points (11 × 11 points on each axis), and the ### Page 25 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 25100 80 60 40 20 0 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 0.1 0.3 0.5 0.7 0.9 albedo s t t = 0.1 t = 0.2 t = 0.4 t = 0.8 t = 1.6 100 80 60 40 20 0 perceived transparency 0.1 0.3 0.5 0.7 0.9 s t 0.1 0.2 0.4 0.8 1.6 100 80 60 40 20 0 0.1 0.2 0.4 0.8 1.6 100 80 60 40 20 0 0.1 0.2 0.4 0.8 1.6 shape: S2 shape: S11shape: S7 extinction coefficient A Bperceived transparencyFigure A1. Results of the experiments using the stimulus set depicted in Figure 1 (for shape S7) and Supplementary Figure S1 (for shapes S2 and S11) to examine the effects of extinction coefficient (σ t ) and albedo (σ s /σ t ) on the perceived transparency of objects. The same results are depicted in 2D format in Figure 2. (A) Average ratings of the perceived transparency (0–99) for nine participants (vertical axis) are plotted against the extinction coefficient (σ t ). Lines with different colors correspond to different albedo (σ s /σ t ). (B) Same data as in A are plotted against the albedo (σ s /σ t ). Lines with different colors correspond to different extinction coefficients (σ t ). Error bars indicate SD.results are shown in Figure A2 (A for σ t /α stimulus set, B for σ t /albedo stimulus set). Response magnitude at each point is normalized by the maximum response and shown by different shades of pink. Coordinates of the peak response are shown in each interpolated map by an asterisk in A and B. We also drew response contours by connecting the points where the interpolated responses are the same (0.1, 0.3, 0.5, 0.7, and 0.9 of the peak response). Coordinates of the peak responses of the neurons exhibiting selectivity to each stimulus set are shown in Figures 9A, 9B. Response contours where the response magnitude was 0.5 of the maximum (50% response contour indicated in red) of the selective neurons are shown in Figures 9C, 9D. ### Page 26 Journal of Vision (2026) 26(5):4, 1–26 Baba, Kanari, Iwasaki, & Komatsu 26extinction coefficient (t)0.1 0.4 1.6 0.1 0.5 0.9 albedo 1 0.8 0.6 0.4 0.2 0 0.9 0.7 0.5 0.3 0.10.1 0.4 1.6 0.3 0.055 0.01 surface roughness () 1 0.8 0.6 0.4 0.2 0 0.9 0.7 0.5 0.3 0.1 -0.10.10.1 0.3 0.5 0.7 0.9 0.9 0.9 0.7 0.5 0.3 0.1A Bstimulus set t /albedo stimulus set unit 1 * *Figure A2. Interpolated and normalized responses of a neuron (Unit 1) depicted in Figure 6 to the σ t /α stimulus set (A) and theσ t /albedo stimulus set (B). Response magnitude is shown by different shades of pink. Lines indicate response contours drawn by connecting the points where the interpolated responses are the same (0.1, 0.3, 0.5, 0.7, and 0.9 of the peak response).