A local neural implementation of histogram equalization

Histogram equalization is a technique for optimizing the utilization of quantization levels-typically in connection with the processing of visual images. A generalized version of histogram equalization as a model for contextual processing in neural networks is advanced. The concept of histogram equalization as applied to image grey scales is reviewed. This idea is extended to more general types of quantization, including domains in which the statistics are gathered temporally as well as spatially. An artificial neural circuit which implements generalized histogram equalization is presented and analyzed. By noting that the feedback paths in this implementation are local, the elements of the author's formal circuit are related to components of a biological neural model. It is concluded that histogram equalization is a useful model for short-term adaptation, that it is easily implemented in artificial neural circuitry, and that it is biologically plausible as well.<>