Discrimination of Density and Clustering on Four Versions of a Stochastic Display

Abstract

A number of studies have used stochastically generated visual displays to study detection and discrimination of differences in pattern texture. As in this study, the texture is controlled by a Markov chain paradigm in one or more dimensions. Texture differences are generated either within a given pattern, or between two patterns which are presented simultaneously or sequentially. Other parameters of the display are held constant (e.g., positive and negative images) and therefore not explored. This study investigates the interaction of two texture variables, density and clustering, with the display parameters of positive/negative image and adjancency/separation of texture images. Stimulus display patterns were generated using a two dimensional Markov process to control density and clustering, and were put on 35mm film. Stimuli were presented on a rear projection display, and subjects were required to determine the sameness or difference of the two halves of the display. The major findings of the study are: (1) Performance on the four versions of the display are not equivalent and differ significantly in some respects; (2) Performance on detecting differences in density and in clustering are consistent with earlier studies; and (3) A sharp discontinuity in performance level as a function of the magnitude of standard/variable difference is present for small differences. The findings of this study raise questions as to the generalization of texture data among similar but not identical displays. The presence of the low level discontinuity is also of interest in the interpretation of this data in terms of the theory of signal detectability and the use of the classical ogive. From the application viewpoint, the non-equivalence of the display variants and the presence of the discontinuity may be of significance to problems of target detection on sonar and radar displays.