A CNN-driven locally adaptive CMOS image sensor

Abstract

A bioinspired model for mixed-signal array mimics the way in which images are processed in the visual pathway. Focal-plane processing of images permits local adaptation of photoreceptor structures in silicon. Beyond simple resistive grid filtering, nonlinear and anisotropic diffusion can be programmed in this CNN chip. This paper presents the local circuitry for sensors adaptation based on the mixed-signal VLSI parallel processing infrastructure in CMOS.