An experimental investigation in the use of color in computational stereopsis

Three different models of incorporating color in computational stereopsis were investigated. The simplest model, the three-channel model, treats color as if it were composed of separate entities. Each color channel is processed individually by the same stereopsis module as used in the achromatic model. The second model, the trichromatic model, treats color as an attribute of features (i.e., edges). Matching is done by comparing these attributes. The third model, the opponent color model, is based on the human vision system. In this model, the red-green-blue colors are combined into three opponent channels before further processing. These models along with the achromatic stereopsis model were implanted and tested with many stereograms to determine the performance of different combinations of color and stereopsis. Experimental results indicate that color information increases performance, communication between color channels during matching is very important, and the overall performance of the opponent model appears to be the best. Also described is the relaxation surface smoothing algorithm, which corrects assignment errors