Self-Adaptive Retinal Processes and Their Implications to Adaptive Control

Abstract

An analysis of two-dimensional optical images (or spatial distributions of other physical quantities) with various signal-to-noise ratios requires an optimum balance between integration (or averaging) to detect weak signals, and differentiation to increase contrasts. This is connected with sensitivity-selectivity balance.

Based on some physiological observations this paper provides certain models of retinal receptive fields which adapt to various conditions. These models called adaptive centrio operators have a relatively small number of inputs. Nevertheless, a mapping system containing them achieves optimization in picture processing for various conditions.