CLEF'12 Proceedings of the Third international conference on Information Access Evaluation: multilinguality, multimodality, and visual analytics
| Hi-index | 0.00 |
Query by image example is still a challenge in image retrieval. The goal of similarity retrieval in images is to get the similar images quickly and accurately in high-dimension space. The accuracy of similarity retrieval in high-dimension space is mainly decided by the features representing images and the method used for similar calculation. Our main goal in this paper is to improve the retrieval speed without great lost of accuracy. We propose a filtering method to greatly reduce the search range based on two assumptions: (1) the similar images will have similar amount of SIFT (scale invariant feature transform) features;(2) the similar images will all contain the important features. In contrast to prior work on similarity retrieval in high-dimension space, we use the distribution of features of images to filter the target images. Experimental results show that our approach can significantly reduce the time complexity.