Depth, motion, color, texture are primary outputs of human visual cortices. Estimating these features are also the desired attributes of a general computer vision system. Many mathematical or empirical methods, although popular in the computer vision field, are not yet as robust as the human vision system. Therefore, it has become an important trend to build visual information processing machines based on knowledge from biological vision system. This thesis aims to build biological stereo vision model based on the extensive studies focusing on primate stereo vision system. The model should be able to extract depth from the two retinal images, as well as to provide new insights into neural system underling the primate stereo vision. Our model assumes that the correspondence problem, critical in stereopsis, is largely solved by a neural network simulating Markov random field (MRF). The neural dynamics of the neural network is assumed to implement the belief propagation algorithm. There are two differences between our proposed model and other stereo vision models based on MRF in computer vision field. First, the likelihood function in our model is constructed on the basis of the disparity energy model because complex cells are considered as front-end disparity encoders in the visual pathway. In addition, our likelihood function is also relevant to several psychological findings. The potential function in our model is constrained by the psychological finding that the strength of the cooperative interaction minimizing relative disparity decreases as the separation between stimuli increases. Our model is tested on three kinds of stereo images. In simulations on images with repetitive patterns, it is demonstrated that our model could account for the human iv depth percepts which were previously explained by the second-order mechanism. In simulations on random dot stereograms (RDS) and natural scene images, it is demonstrated that false matches introduced by the disparity energy model can be reliably removed by our model. A comparison with the coarse-to-fine model, a well-known model in the literature, shows that our model is able to compute the absolute disparity of small objects with larger relative disparity. Our model is also related to several physiological findings. The hypothesized neurons of the model are selective for absolute disparity and have facilitative extra receptive field. There are plenty of such neurons in the visual cortex. In conclusion, s...
修改评论