Texture discrimination based on neural dynamics of visual perception

A basic problem in computer vision is to develop texture discrimination techniques that perform as well the human visual system. A theoretical framework that relates dynamics of texture discrimination over context changes with perceptual discrimination is presented. The framework consists of a texture space, a perceptual space and a representational space. Texture stimuli are mapped on to the representational space. A nonlinear psychophysical dynamic transformation is used to map texture stimuli input to responses. This dynamics couples the environment to an internal feedback loop that models the non-linearities in sensory and neural processes of psychovisual perception. The framework can be [Coners, R.W., Ng, C.T., 1989] used to perform texture discrimination that preserves perceptual similarity with the performance of the visual system and also used to study how discriminability of texture patterns is affected by changes in a particular texture property over time. Using this framework a procedure to model discrimination of texture is presented. The nonlinear psychophysical dynamics are implemented using simple finite difference scheme and can be easily converted to parallel algorithm.