A Biological Stereo Vision Model Based on Markov Random Field

Human can obtain 3D percept from 2D images on the retinas. However, the neural mechanism of biological stereo vision is largely unknown. On the other hand, Markov random field and belief propagation algorithm have given birth to stereo algorithms with top performance. I propose that brain may employ similar neural network to perceive depth. I design a model combining Markov network with known physiological and psychological findings. Complex cells are considered as disparity encoders in brain, and their response can be predicted by the disparity energy model. In my model the likelihood function (also known as the data cost function) is based on response of complex cells. Furthermore, the likelihood function has connection to several psychological findings. The model is tested on two kinds of stereo images and compared with other models. The simulation demonstrates that Markov network can solve ambiguities in responses of complex cells quite well. This result lays the computational foundation for my proposal.