The RANSAC (random sampling consensus) algorithm is an estimation method that can obtain the optimal model in samples containing a lot of abnormal data. This algorithm uses a small number of points in the data to estimate the model, and then uses the remaining points to check the model. It is now widely used in computer vision image stitching. This algorithm has disadvantages such as slow operation speed and poor sample adaptability. In order to improve the efficiency of image registration, many researchers have made improvements on the basis of the RANSAC principle. This article introduces the principles and shortcomings of RANSAC, and introduces four ways to improve it in view of its shortcomings. The advantages and problems of the improved algorithm are analyzed to provide a basis for the field of image registration.