A sift-based method for image mosaic

Abstract

This paper proposes a mosaic algorithm based on SIFT(Scale Invariant Feature Trasform). Keypoints and their descriptors are exacted by SIFT, which ensures the algorithm robust to translation, rotation, noise and scalling. Then proposed matching algorithm is used to match the keypoints. First, rough match pairs are obtained by Nearest Neighbor algorithm. Second, match rate information of the neighbor keypoints and the global information are calculated to update the match precision, so each keypoint matching rate spreads to its neighborhood, which eliminates a large number of mismatches to achieve the exact match. Third, RANSAC is applied to obtain the transformation matrix. Finally, two images are stitched by linear weighted fusion algorithm. The experiment results confirm the feasibility and improvement of our model.