Robust Weighted Keypoint Matching Algorithm for Image Retrieval

The important factor which controls the accuracy of image matching process is a keypoint extraction. This paper describes a new weighted keypoint matching algorithm for improving the performance of image retrieval. The basic concept is to use the edge information. First, we compute the Scale Invariant Feature Transform (SIFT) features as keypoints from an input image. Also, the edge image is obtained for updating the ranking of the extracted keypoints. With the obtained edge map, we assign weights to keypoints where they correspond on the edge map. Finally the ranking of keypoints is re-ordered and Top-32 keypoints are selected based on the RootSIFT to matching process. Through the experiments, we verify that the proposed algorithm achieves 2.4% of more accuracy than the original SIFT detector when various distortions exist.