Edge Detection Model based on Involuntary Eye Movements of the Eye-retina System

Recent results in retinal research have shown that ganglion receptive fields cover the mammalian retina in a mosaic arrangement with very small overlap. Classical biology inspired models of the retina performing image processing operations such as edge-detection, in contrast, by using convolution of a filter and the image, map overlapping input image regions to output signals. Without such an overlap, the output becomes a sparse sampling of the filtered output, which leads to the loss of important pieces of visual information. This paper proposes a retinal model with a non-overlapped receptive field structure. The model implements involuntary eye movements, tremors and drifts. Artificial eye-tremors, small vibrations of the eyes, achieve receptive field overlaps through time. Artificial eye-drifts are also implemented and used for orientation selective edge detection. Based on the experimental results, the following three hypotheses are claimed in this paper: 1) convolution-based retinal models are biologically inadequate, 2) eye tremors may play an important role in retinal contour perception, and 3) artificial eye drifts may also be applied in artificial vision systems to achieve orientation selectivity.