Adaptive response organizer network for space-time patterns in low level vision.

Abstract

In this paper we present a biologically-inspired generalized artificial model of retinal neural response organization that can serve as a useful framework for analysis of response dynamics in artificial neural networks, specially in the visual systems. Adaptive Response Organizer (ARO), is an ON-centre OFF-surround class of artificial neural network models for feature abstraction and object identification under varying background conditions. The proposed architecture utilizes the existing literature on retinal functionality to develop a self-organized scheme to form visually evoked activation maps in response to arbitrary sequences of analog (gray scale) visual input patterns. It is conceptually organized into three processing layers, where visual information flows vertically through the layers and reorganized laterally by interneuronal action. The temporal response configuration of visually responsive neurons acquire the antagonistic centre-surround characteristics to promote detection of weak contrasts and rapid changes to transient and steady state stimulus at early stages of vision, for optical scene interpretation.