Abstract
Many real-world materials are translucent (e.g. skin), but our understanding of translucency is very limited. Previous findings of human perception of translucency are mainly based on rendered images, which are restricted by their diversity of appearance and realism. We measure observers' translucency perception with photographs of real objects. Here, we examine individual differences and the role of color in perceiving translucent appearance through a classification task. Observers had unlimited time to classify 250 images as “Translucent”, “Opaque” or “Unsure”. Stimuli are photographs of real objects made of natural or artificial materials, such as fruits, wax, ceramics. In Experiment 1, the stimuli are presented as RGB images. In Experiment 2, the images were converted to grayscale by extracting the luminance channel. Twenty-five observers completed each experiment. First, we found substantial individual differences in both color and grayscale conditions. In the color condition, only 22% images reached 96% agreement among observers wherein grayscale condition 24% images reached 96% agreement. Furthermore, more observers reported “Unsure” in the grayscale condition in comparison to the color one. We constructed the representational dissimilarity matrices (RDMs) based on individual votes for each image pair for both conditions. The patterns of the image-by-image RDMs are visually different between the color and grayscale conditions (Kendall's τA=0.27, p<0.01), revealing a shift in observers’ responses to certain stimuli and a significant but weak correlation between the two RDMs. For example, some objects (e.g. tennis ball) uniformly voted as “Opaque” in the color conditions were voted by some observers as “Translucent” in the grayscale conditions, thus, resulting in more individual difference. Our results suggest that there are substantial individual differences in translucency classification from photographs of real objects, likely due to interpersonal differences in interpretation of translucency and object categories. Moreover, removing color has an effect on the votes for some stimuli.