Image Enhancement Directed by Maximum Contrast Gradients Encoded by a Lattice of Receptive Fields

The work is focused on the problem of image restoration (decoding) after perceptually meaningful compression (encoding), based on the periphery neuromorphic model, inspired by the mechanisms of human visual system. The modeling of the retinal mechanisms is carried out in the current work on the basis of the most realistic representation of input data in the form of a stream of events (counts), triggered by photoreceptors. The statistical description of the streaming data, chosen in the form of a Poisson two-dimensional point process and called the sampling representation, was obtained and substantiated in previous papers. To adequately model the retinal encoding process the sampling representation is equipped by the special structure, motivated by the concept of receptive fields. This structure implements a number of well-known neural mechanisms, including center / surround inhibition. Decoding issues are considered in frames of input data spatial contrast restoration, on the basis of the primary visual cortex simple cells response model. It is shown that the model of coupled ON-OFF encoding of input counts allows to restore sharp image details in the form of local edges. To justify the adequacy of the synthesized encoding procedure, at the end of the work, we demonstrate an example of edge-directed image enhancement, fulfilled by edge delineation.