Image representation in visual cortex and high nonlinear approximation

Abstract

We briefly review the "sparse coding" principle employed in the sensory information processing system of mammals and focus on the phenomenon that such principle is realized through over-complete representation strategy in primary sensory cortical areas (V1). Considering the lack of quantitative analysis of how many gains in sparsenality the over-complete representation strategy brings in neuroscience, in this paper, we give a quantitative analysis from the viewpoint of nonlinear approximation. The result shows that the over-complete strategy can provide sparser representation than the complete strategy. © Springer-Verlag Berlin Heidelberg 2005.