Extraction of three-dimensional scene from the stereo images is the most effective research area in the field of computer vision. Stereo vision constructs the actual three-dimensional scene from two stereo images having different viewpoints. Stereo matching is a correspondence problem, that means it ascertains which part of image corresponds to which part of another image ,where variations inside two images is due to the movement of camera or elapse of time. Many stereo matching algorithms have been developed in order to construct the accurate disparity map. This paper presents a review on various stereo matching techniques. The comparison among existing techniques has clearly shown that none perform optimistically every time. This review has shown that the existing methods in stereo matching involve median filtering. But median filter is not effective for high density of noise. Besides mean-shift segmentation is being used for disparity refinement in existing methods, which can be enhanced by using improved mean-shift segmentation, available now days. In addition, guided filter has been used by many algorithms, but this can be replaced by joint trilateral filters.