Towards photosensor movement-adaptive image analysis in an electronic retinal prosthesis

Abstract

By way of extracellular, stimulating electrodes, electronic vision prosthesis aims to render discrete light spots--so-called phosphenes--in the visual field, thereby providing a phosphene image serving as a rudimentary remediation of profound blindness. It is proposed that a digital camera, or some other photosensitive array, captures frames, the frames be analyzed, and phosphenes be actuated accordingly. We present a numerical experiment wherein we observed the phosphene image in response to a set of stimuli for various image analysis schemes. We used the mutual-information function to quantify the efficacy of analysis schemes; the function penalizes a scheme for introducing redundancy to the phosphene image, while accounting for the probability of each stimulus. We demonstrate an effective scheme involving Laplacian of Gaussian (nabla(2)G) kernels geometrically transformed in accordance with phosphene layout. Further, we propose adapting the kernels comprising a scheme in accordance with photosensor movement.