A Fast Implicit Algorithm of Feature Matching Based on Space Segmentation Technology

Scale Invariant Feature Transform (SIFT) is now the most efiective and widely used technique for feature extract. Although it has great ability in content description, there are still some problems in its feature matching. Generally, SIFT and correlative algorithms compute distance and use neighbor algorithm to look for the optimal matching couples. The disadvantages of such way are such like high complexity, instability of matching couples and so on. Especially, when huge amount of images need to be retrieved or recognized, its matching e‐ciency is very low. To solve this problem, a new matching way based on feature space division under multi-resolution is proposed in the paper. Through the algorithm, the feature space is divided into several sub-blocks under difierent resolutions. And then each sub-block is assigned to an only code. So that those feature points which are located in somewhere can be represented by the codes. At last, the feature matching can be completed through code matching, which is somehow easier to do. The experiments show that this algorithm can improve the matching e‐ciency greatly when the matching accuracy is kept as well.