Abstract
Chemical and physical properties of objects provide them with specific surface patterns of colors and 3-D textures. Endogenous and exogenous forces alter these colors and patterns over time. The ability to identify these changes can have great utility in judging the state and history of objects. To evaluate the role of color and 3-D texture cues, we used calibrated images acquired from 15 different viewpoints of 26 real materials undergoing changes (Courtesy of Shree Nayar and Jinwei Gu). Materials included fruits, foods, woods, minerals, metals, fabrics and papers, and changes included drying, burning, decaying, rusting, oxidizing and heating. Observers were asked to identify materials and types of changes for color and gray-scale images. Observers obtained 3-D information by varying the viewing angle of the image (deformation of the frame provided estimates of the slant and tilt of the material with respect to the observer). The images were shown in three sets of trials: one image of the surface, two images of the same surface at the beginning and end of a natural change, and image sequences of the time-varying appearance (number of time samples varied from 10 to 36). The presence of color cues improved performance in all conditions but most dramatically in the organic category. This may be because certain color patterns occur only in organic fruits and vegetables. Identification of materials improved if observers saw two states of the material, but the complete image sequence did not improve performance if images were restricted to fronto-parallel view-points. The ability to examine the material surface from several viewpoints improved performance, thus showing the importance of the 3-D micro-structure of the surface texture. The role of color in object recognition has been controversial, but this controversy may be resolved as color's role in material perception becomes clearer.
Grants EY07556 & EY13312 to QZ.