Research On The Cognitive Stereo Vision

As one branch of brain science and cognitive science, the main research work of cognitive is to explore within the brain to find when and where the information is processed. The people who engage in pattern recognition are more concerned about how brain processes visual information; they set up a cognitive based machine vision in the visual cortex of disparity detection and eye movements with the latest research results of neurophysiology. But the two algorithms are not integrated so far.This thesis researched the above questions and proposed new methods with the previous search results. The experiments show the effectiveness of the proposal. The thesis includes the following aspects.(1) Research the cortical disparity detection mechanism and the visual attention mechanism in a single framework. The goal of the cortical disparity detection mechanism is to obtain three-dimensional environment information from two-dimensional images. The goal of the visual attention mechanism is to extract and transfer visual focuses from mass visual information with the color, shape, and other information. This thesis first analyzes the intrinsic link and relations between the two algorithms and gets a result that they should be researched together. Then a bottom-to-up module is used to extract meaningful objects information from color, direction and intensity and feedback the results to visual disparity detection module and guide it.(2) Propose a self-confidence point spread algorithm in the visual cortex parallax detection mechanism. The result supports the disparity energy model that there exists both location and phase disparity in the visual cortex parallax detection mechanism. The translation is simulated to eye movements in this mechanism and one-dimensional Gabor filters are used to compute the phase disparity. The disparity map is generated with the probability. The visual cortex parallax detection mechanism and the constants in traditional parallax theory are combined in this thesis. A self-confidence point spread algorithm is merged into the visual cortex parallax detection mechanism to correct the disparity map. Experiment result shows that the proposed algorithm is effective and could reduce the mismatch.(3) Research the eye movements in visual attention. There was no constrain or restriction to eye movements in the traditional visual cortex disparity detection mechanism, so the traditional image translation covers the entire range of possible disparity. This method does not use the brain so that it brings mass computation and useless information. This thesis uses the results to control the eye movements and reduces computation of the match. (4) Use BPA algorithm to reconstruction in three-dimensional scene. BPA is an algorithm to build normalizes triangle mesh in 3D space. BPA is widely used in human face reconstruction with its efficiency and robustness. The thesis extends the algorithm to scene reconstruction. The result shows that it is reliable to use this algorithm in scene reconstruction and can satisfy this need.