Abstract Visual estimation of material properties, such as gloss, viscosity, and wetness, is one of the main tasks of human Shitsukan recognition. It is computationally challenging since the image formation involves complex optical interplays of
material, geometry, and illumination, and because physical material properties are
inherently complex and high-dimensional. It seems difficult for human vision to estimate material characteristics from a single image in a strict inverse-optics way. The
hypothesis that human vision heuristically uses image-computable simple features
for estimating material properties (e.g., the skewness of the luminance histogram
for gloss estimation) was proposed but has been criticized for its ignorance of other
relevant stimulus factors (e.g., 3D shape). The complexity of human material computation remains a controversial issue, and the recent success of artificial neural networks in explaining human vision has amplified expectations about the complexity
of the underlying neural computation. Here, we argue that specifying relevant image
features for material perception still is a useful strategy for an explicit understanding
of the human visual computation of material, as long as the study also analyzes the
underlying physical mechanism and explains why these features work and when they
fail to predict human material perception. Furthermore, for a full understanding of
the human material processing, we should investigate how the material affects the
perception of the other attributes, such as shape and motion, in addition to how the
material perception is affected by the other attributes.