Abstract

Victor has developed binary “random-plaid” patterns he proposed to have random statistics up to third order but that are highly discriminable from random binary noise. By concatenating blocks of random-plaid patterns and noise I have generated patterns with random statistics up to seventh and 15th orders that are also discriminable from noise. The question is, what does this tell us about visual processing? This was addressed by computing the autocorrelation functions (ACFs) of random-plaid patterns to encapsulate their second-order statistics. The ACF of a piece of random-plaid pattern was not similar to that for a noise pattern of the same size; it had a much higher variance and also exhibited a random-plaid structure. The statistics would converge only for patterns of infinite extent. This raises questions about the utility of the Victor approach to pattern analysis.

© 1992 Optical Society of America