Iterative Orientation Tuning of Simple Cells in VI: A Comparative Study of Two Computational Models for Contrast Detection in Images

摘要

The orientation selectivity of simple cells in visual cortex gives a striking example of the biological system perfectly adapted to the perception of oriented stimuli. Several models have employed major principles of orientation selectivity for the processing of contrast variations in images. We have recently suggested a model for iterative orientation tuning, in which the astonishingly regular layout of simple cells is explicitly involved in the processing of oriented stimuli. In this work. We extended the iterative model by incorporation a mechanism of cross-oriented inhibition. We then investigated the two models using synthetic, noisy and natural images. We found that the two models account for a large fraction of the contrast invariance of orientation selectivity another striking aspect of the behavior of simple cells. Our results indicate that the iterative processing of visual stimuli combined with local amplification of proximate simple cells is responsible for ～75% of the contrast invariance Contrary to some earlier studies, the cross-oriented inhibition did not have any significant contribution to the contrast invariance but accelerated the convergence of the iterative processing on a stable solution. When probed with different images, the new model with cross-oriented inhibition generated a clear pattern of object contours.