Lessons from biological processing of image texture

Abstract

When designing artificial vision systems, it may be useful to examine the solutions 0.5 billion years of biological evolution have produced. Recent studies of human vision; studies of macaque visual cortical function; and behavioural studies of bee vision, all indicate that different species have evolved related approaches for discriminating image textures. This common strategy uses short-range 4th-order spatial correlations. Isotrigon textures, ensemble averages of which have 3rd-order correlation functions that are equal to 0, are useful for studying this sense. Recent results from humans and bees, and methods for producing new isotrigon textures are described.