Aiming at the poor real-time and robustness of registration method for outdoor augmented reality while mobile devices was used, a new image registration method based on improved speed up robust features (SURF) algorithm and Kanade-Lucas-Tomasi (KLT) algorithm in the mobile augmented reality is proposed. The method improves the SURF algorithm by combining the fast retina keypoint (FREAK) algorithm which has a very good performance of extracting descriptors, expressed in speed, quality and quantity, SUFREAK algorithm is formed. It makes up for the shortcoming of slow feature points extraction in SURF algorithm. SUFREAK algorithm improves the descriptor extraction efficiency and maintains the better robustness of the algorithm at the same time. Using the correlation between the close frames of video, the KLT optical flow tracking algorithm is used to track and predict the position of natural feature points in the open air, it will be effective to improving the real-time performance of image registration method. The experimental results show that the improved SURF algorithm has higher real-time performance and robustness under complex outdoor environment conditions, and the registration method of mobile augmented reality based on improved SURF and KLT algorithm has a good real-time performance of system and the efficiency of image recognition has been greatly improved.