Abstract
Material properties, 3D shape and illumination conditions all contribute to the 2D (retinal) image. All three ingredients of the image formation process have previously been investigated, but we were interested in combining 3D shape perception with image information to infer material and light properties.
We selected 16 painted and photographed textile samples which showed a variety of draped fabrics. For each picture, we selected 20 sampling locations where we measured the RGB color which was converted to a (relative) luminance value. These same locations were used in a gauge figure task where we used an attitude probe to quantify the perceived local slant and tilt. For this task, 8 observers participated per textile in an online experiment.
If reflectance was Lambertian and the light conditions were collimated in frontal direction, one would expect a negative correlation between slant and image luminance. Although we do not assume light conditions or reflectance properties a priori, we explored whether slant/luminance correlations were related to the material class.
Correlations varied between -0.5 to +0.9 and materials classes seemed to covary distinctively with these values. Negative correlations were primarily found with matte materials. Stimuli with correlations around 0 (between -0.2 and 0.3) looked like shiny satin materials. The large positive correlations were found with hairy, velvety looking materials. We did not find differences between painted and photographed textiles.
Although a more complex model may reveal more accurate insights into the depicted material properties, this simple model seems to relate slant/illumination information to materials class quite well.