Aiming to solve the problem of low accuracy and poor stability of random sample consensus (RANSAC) algorithm in feature point matching, an image registration algorithm based on smoothness constraint and cluster analysis is proposed in this paper. First, the scale information and spatial angle order of neighborhood matching feature points are used to construct a smoothness constraint, and the initial matching points are divided into a sampling set with a high inlier rate and a verification set with a high inlier number. Then, the solution is solved by repeated sampling and model testing. Next, the inlier set is temporarily determined, and cluster analysis is performed on it. Further, the optimal inlier set is selected according to the distribution quality of the cluster center in the image overlapping area. Finally, the optimal inlier set is used to solve the model parameters to achieve image robust registration. The simulation results show that compared with the RANSAC algorithm, the registration accuracy of the algorithm improved by 26.83%, and the error standard deviation is reduced from 0.68 to 0.19.