A Fractional Stereo Matching Based On Color Segmentation And Census Transform

Stereo matching technology is the key of three-dimensional reconstruction, its precision directly determines the result of three-dimensional reconstruction. However,stereo matching faced with the problem of being difficult to meet accuracy and speed requirements simultaneously, and being sensitive to noise and the illumination variance.At the same time, it is very difficult to quickly obtain high-precision disparity map. Stereo matching is a basic algorithm and key problem, and it becomes a hot research topic in the field of measurement at home and aboard. In this paper, Aiming at the disadvantage of stereo matching method, looking its disparity searching strategy and similarity measurement function as the center, and on the basis of stereo matching method’s research,color segmentation and the non-parameter transformation is introduced in the algorithm,this paper proposes a fast and robust fractional stereo matching algorithm.Firstly, in the view of stereo matching disparity searching strategy, introduce color segmentation to the stereo matching method, a parallax constraint based on color segmentation is proposed, and adopt a fractional search strategy of disparity to a certain range disparity. That is to say, It adopt mean shift algorithm to segment images, and calculate the first matching cost based on the support region with arbitrary shape and size,then form the parallax constraint. Then according to the result of disparity constraints,calculate the second dense disparity map. This fractional matching method based on color segmentation not only obtains the higher accuracy of disparity map, but also greatly improves the matching speed.Secondly, in the view of stereo matching method’s similarity measurement function,aiming at the limitations of being sensitive to noise and the variance of light, and on the basis of improving the traditional non-parametric transformation, a stereo matching algorithm based on fusion mechanism of similarity measurement function is proposed.That is to say, one matching cost based on support region with arbitrary shape and size is obtained by using color segmentation information, then another matching cost of weighted non-parametric transform in similar color region is obtained, finally, joint matching cost isobtained by the fusion of the both matching cost. This fusion mechanism of joint matching cost function not only reserves the advantage of being robust to illumination variance of traditional non-parametric transformation, but also improves the accuracy of the disparity matching.Finally, several experiments and evaluation data show that, the stereo matching algorithm can meet accuracy and speed requirements simultaneously, it is also robust to noise and the variance of light. So the method is effective, and it is a right stereo matching method.