サンファ リー, ジョンイル パーク, 井上誠喜
A New Stereo Matching Algorithm
Based on Bayesian Model
Abstract:In this paper, the general formula of Bayesian model for stereo matching algorithm is derived and implemented with simplified probabilistic models. The probabilistic models are independence property between the neighborhood disparities in the configuration, and similarity of disparities in adjacent neighborhood. The formula is the generalization of Bayesian model of stereo matching, and can be changed into the some forms of Bayesian model of stereo matching according to the probabilistic models in the disparity neighborhood system or configuration. And, this paper performed two kinds of experiments. One is the comparison of the performance between the proposed algorithm and the other algorithms, such as the conventional Bayesian algorithm [?, ?] and block-based squared sum algorithm. The other experiment applies the multiple view stereo images to the proposed algorithm of Bayesian model. According to the experimental results, we can conclude the following facts. The first is that the derived formula is the general form and can be changed into the some different forms based on the reasonable probabilistic assumptions. The more accurate is the assumed probability model, the better disparity map can be generated with this formula. And, this Bayesian model can be developed with various probabilistic model and configuration. The second is that it is very important to generate the initial energy space in Bayesian model of stereo matching, so the multiple stereo images are useful for the estimation of the better disparity map.