HISTOGRAM CONTRAST ANALYSIS AND THE VISUAL SEGREGATION OF IID TEXTURES

A new psychophysical methodology is introduced, histogram contrast analysis, that allows one to measure stimulus transformations, f, used by the visual system to draw distinctions between different image regions. The method involves the discrimination of images constructed by selecting texture micropatterns randomly and independently (across locations) on the basis of a given micropattern histogram. Different components of f are measured by use of different component functions to modulate the micropattern histogram until the resulting textures are discriminable. When no discrimination threshold can be obtained for a given modulating component function, a second titration technique may be used to measure the contribution of that component to f. The method includes several strong tests of its own assumptions. An example is given of the method applied to visual textures composed of small, uniform squares with randomly chosen gray levels. In particular, for a fixed mean gray level mu and a fixed gray-level variance sigma2, histogram contrast analysis is used to establish that the class S of all textures composed of small squares with jointly independent, identically distributed gray levels with mean mu and variance sigma2 is perceptually elementary in the following sense: there exists a single, real-valued function f(S) of gray level, such that two textures I and J in S are discriminable only if the average value of f(S) applied to the gray levels in I is significantly different from the average value of f(S) applied to the gray levels in J. Finally, histogram contrast analysis is used to obtain a seventh-order polynomial approximation of f(S).