Visual discrimination of optical material properties: a large-scale study

For elucidation of complex visual processing such as material perception of realistic objects, multiple research approaches should be combined. Shearing stimulus and response data is an effective strategy to make the results of different studies directly comparable, and to assist a wide range of researchers with different backgrounds to jump into the field. We therefore constructed a database of a variety of material images with standard human discrimination performance attached. We created various material images by using physically-based computer graphics techniques and conducted psychophysical experiments both in the laboratory and through crowdsourcing. The observer’s task was material discrimination for six material dimensions (gloss contrast, gloss sharpness, translucent vs. opaque, gold vs. plastic yellow, glass vs. silver, and glossy vs. painted) with several task difficulties controlled by material parameters. We also varied illumination consistency and object geometry. We used a non-verbal procedure (oddity task) for our database to be widely used in diverse studies such as cross-cultural, cross-species, clinical and developmental studies. Results showed that discrimination performance was affected by illumination condition and object geometry, in agreement with previous gloss perception studies, although the pattern of effects was slightly different for some material dimensions. Specifically, we found that discrimination of translucency was easier for rugged shapes than for smooth shapes, while the opposite was the case for gloss perception. We also found that the ability to discriminate the spatial consistency of specular highlights on glossiness perception had a larger individual difference than other tasks. The performance obtained through crowdsourcing was highly correlated with that in the laboratory, which suggests that our database can be used even when the experimental conditions are not strictly controlled under non-laboratory environments. Several projects using our dataset are underway.