Decoding Neuromorphic Codes of Images in the Marr’s Paradigm

The work is devoted to the results of analysis/synthesis of image encoding based on the periphery neuromorphic model inspired by the mechanisms of human visual system. To adequately model the retinal mechanisms analysis and synthesis are carried out on the basis of the most realistic representation of input data—external stimuli—in the form of a stream of triggered receptor events. The statistical data model is chosen in the form of a Poisson random counts, the rationale for which was obtained in our previous papers. On this basis a statistical description of the input data in the form of a sampling representation is accepted. The model of retinal coding of input data is considered within the framework of the concept of receptive fields. This coding model implements a number of well-known neural mechanisms, including central/lateral inhibition. Decoding issues are considered in the context of reconstructing spatial contrasts of the image, partially modeling the responses of simple cells of the primary visual cortex. It is shown that the model of coupled ON-OFF encoding allows to restore sharp image details in the form of local edges. At the end of the article the interesting connection with the Marr’s theory of computer vision is discussed.