Objective Feature-based registration algorithms have significant advantages such as strong robustness and good pertinence, and is widely used in the field of image registration. However, the accuracy of such method is greatly affected by feature construction between images and environmental noise. This study aims to improve its shortcomings. Methods Based on SURF and ORB algorithms, the SURF-ORB algorithm is proposed. The Moving Image and the Fixed Image are divided into two parts for registration. In the registration process, the Harris response value of the image feature points extracted by SURF is first optimized, and the gray centroid method is used to determine the main direction of the feature points. Then, calculate the rBRIEF (rotated BRIEF) descriptor, and use the Hamming distance to match the feature points. Finally, RANSAC fine matching algorithm is added to eliminate false matching points. Results and conclusion In this study, the registration results, anti-noise ability and multimodal registration ability of SURF, ORB and SURF-ORB algorithms were compared and analyzed, and verified the high registration accuracy, registration speed and anti-noise ability of SURF-ORB algorithm. The innovation of the article The study first combines SURF and ORB algorithms and applies them to brain cross-sectional images.