Integrated Visual Saliency Based Local Feature Selection for Image Retrieval

nowadays, local features are widely used for content-based image retrieval. Effective feature selection is very important for the improvement of retrieval performance. Among various local feature extraction methods, Scale Invariant Feature Transform (SIFT) has been proven to be the most robust local invariant feature descriptor. However, the algorithm often generates hundreds of thousands of features per image, which has seriously affected the application of SIFT in content-based image retrieval. Therefore, this paper addresses this problem and proposes a novel method to select salient and distinctive local features using integrated visual saliency analysis. Based on our method, all of the SIFT features in an image are ranked with their integrated visual saliency, and only the most distinctive features will be reserved. The experiments demonstrate that the integrated visual saliency analysis based feature selection algorithm provides significant benefits both in retrieval accuracy and speed.