A Stereo Matching Algorithm Guided by Multiple Linear Regression

In the stereo matching,the matching of edge pixels in depth image is one of the challenging problems of this technology.The local stereo matching based on the guided image filtering can protect the edge of the depth image and improve the matching precision and accelerate the speed,but this makes the image halo and also introduces a lot of noise in the edge area of the image.In this paper,ridge regression of guided filter is extended to multiple linear regression,and a stereo matching framework of multiple linear regression is proposed to extend the cost aggregation method.The framework is designed to better protect the edge of the depth image.To improve the matching accuracy of the disparity,the proposed algorithm sets up the multiple regression equation of the image pixel value and the gradient information as the variable.Then a weighted combination of cost aggregated values is carried out with the guidance filtering alone.At the same time,the concept of credibility is defined.It is described by the relationship between the minimum value and the minor value of the cost aggregation to avoid ambiguity when facing the disparity selection.The algorithm in this paper is tested on the Middlebury platform.The results show that the framework can effectively improve the precision and reduce the noise.Compared with some high performance algorithms,the algorithm can get high quality disparity map.