Application of nonlinear texture dynamics for image classification

摘要

Visual processing is a psychophysical process. In this paper, it is shown that texture processing by the human visual system shows nonlinear dynamic patterns. This is done by developing recursion equations that describe the dynamics in the perception and maps input texture patterns to output observables. The recursion variables are complex where the real and imaginary parts signify the signal and noise parts of the perception process, respectively. The output of the dynamics is used to perform texture discrimination between pairs of textures. These results are compared with results from experiments with human subjects. A good correlation is obtained between the results using texture dynamics and that of experimental data. This shows that texture perceptual dynamics is an important ingredient in developing computer algorithms that preserve perceptual similarity with human visual system processing of texture. Based on this an algorithm is developed to perform classification of textures. The algorithm is applied to classify natural and synthetic textures. This algorithm shows an improvement in the classification accuracy over traditional methods.