A SIFT Pruning Algorithm for Efficient Near-Duplicate Image Matching

The number of SIFT features extracted from an image is usually large and cannot be adequately controlled,which usually results in poor system performance of low efficiency and instability.A SIFT pruning algorithm is proposed to address the above issues in this work.The algorithm measured discriminative power of keypoints by combining the weighted contrast and ratio of the principal curvature,then extracted the proper number of most significant keypoints within a given range through a two-phase filter process in the steps of keypoint localization and orientation assignment.The experiments show that the proposed algorithm can effectively control the number of features and provide higher accuracy than the previous pruning algorithm.The experiments also indicate that the proposed pruning algorithm achieves much higher efficiency and stability with a comparable matching accuracy in comparison to the original non-pruning SIFT algorithm.