An analysis/synthesis model for the human visual system based on subspace decomposition and multirate filter bank theory

Abstract

A two-stage model of the human visual system (HVS) comprised of low-level (analysis) and a high-level (synthesis) processing stages is presented. The processing done at the low-level stage is described using two approaches, one based on a subspace decomposition and the other on a multirate filter bank. These two models simultaneously provide the high-level processing stage with different perspectives or representations of the retinal image. Based on these representations, the high-level processing stage attempts to perceive the encoded visual information. A single framework, which models the hypercolumn-based structure in the striate cortex and combines the different retinal image representations, is proposed. The model resolves differences among previously proposed HVS models, and suggests the development of image processing algorithms which combine and exploit the information provided by different transformations or representations of the input image.<>