Image registration is a process of matching and superimposing two or more images obtained under different conditions (time, sensing equipment, climate, angle, etc.). Typically, each registration technique is specific to a particular class of application, with some limitations. An fusion image registration algorithm based on feature is presented in this thesis. Firstly, the corner detection method is combined with wavelet transform and USAN regional features. Then, the feature points are matched by cross correlation and RANSAC(Random Sample Consensus). Finally, the thin-plate spline algorithm is adopted. This method can make up for the shortcomings of different registration methods to improve the accuracy. Through experiments on a large number of images, such as characters and scenes, it is proved that the algorithm has good registration accuracy, environmental adaptability and real-time.