Abstract
Purpose: To determine how many and what kinds of mechanisms are used to discriminate random, achromatic textures. Method: Stimuli were grids of uniform squares with randomly scrambled Weber contrasts. A given such “scramble” H was constrained to conform to a prescribed histogram h. Subjects were required, in brief displays, to localize a rectangle of texture K (with one histogram k) on a background of H (with another histogram h). We assume scrambles are encoded with a small number of mechanisms M, each characterized by an impact function giving M's sensitivity as a function of grid-square contrast. Two scrambles are assumed to be discriminable only if they induce sufficiently different levels of activation in one or more mechanisms. If human vision has N or fewer such mechanisms, then any N+1-dimensional space of histograms must determine nontrivial metamers: indiscriminable scrambles with different histograms. For several subspaces of 4th order polynomial histograms, observers adjusted histogram parameters to achieve minimal discriminability while maintaining a large, fixed Euclidean distance between histograms. If minimally salient scrambles support no better than chance discrimination performance, we take them to be metamers. Results: (1) the space of all scrambles with 4th order polynomial histograms contains nontrivial metamers, but (2) 3D subspaces of this space do not. Conclusion: Human vision has three scramble-sensitive mechanisms. Further analysis of the adjustments made by observers in the various subspaces tested yield 4th-order polynomial approximations of the impact functions of the three mechanisms; one is sensitive primarily to scramble mean luminance (or “brightness”), one to contrast variance (or “energy”), and a third to the density of the blackest texture elements (or “blackshot”).