Abstract: A correction method for feature points mismatching is proposed based on improved shape context, aiming at the mismatching problem of corresponding points needed for the depth estimation for the bifocal image. First, the SSURF feature descriptor is presented, as it has good invariability to changes such as scaling, view angles and illumination. By the use of SSURF feature vector matching, object images which had been obtained based on a monocular stereo vision system are processed in two different focal length images, so corresponding SSURF feature points are obtained. Then, an improved shape context descriptor is proposed to correct the error matching of local feature points. The distance between dimensional target points as well as image points and the center of the image, and the geometric relationship between the camera focal lengths are used to complete the related computation and therefore to obtain the depth information of object by using the corrected feature points. The experimental results show that the correction method introduced has small errors for depth estimation and good effect for estimation.