Image Stitching Based on Improved SURF Algorithm

Abstract

In order to solve the problem of uneven distribution of picture features and stitching of images, an improved SURF feature extraction method is proposed. Image feature extraction and image registration are the core of image stitching, which is directly related to stitching quality. In this paper, a comprehensive and in-depth study of feature-based image registration is carried out, and an improved algorithm is proposed. Firstly, the Heisen detection operator in the SURF algorithm is introduced to realize feature detection, and the features are extracted as much as possible. Secondly, the characteristics are described by BRIEF operator in the ORB algorithm to realize the invariance of the rotation change. Then, the European pull distance is used to complete the similarity calculation, and the KNN algorithm is used to realize the feature rough matching. Finally, the distance threshold is used to remove the matching pair with larger distance, and then the RANSAC algorithm is used to complete the purification. Experiments show that the proposed algorithm has good real-time performance, strong robustness and high accuracy.