Toward evolutionary autonomous agents (ϵA2) for computational perception

The paper introduces a new class of evolutionary autonomous agents (εA²) that evolve in a digital image environment and emerge (or extract) salient image features. The behavioral repository for the proposed autonomous agents consists of self-reproduction, diffusion, mutation-based search, decay, and cease to exist, each of which is triggered and hence determined by the specific local stimuli of the two-dimensional image environment. The proposed scheme of reproduction-and-diffusion through artificial-environmental selection, when employed as a method of evolutionary computation in image processing, can readily manifest both interesting fixed-point properties and robust behavior capable of dynamically adapting to different digital image environments