Bag-of-Visual-Words Based Object Retrieval with E~2LSH and Query Expansion

摘要

Boosted by the invention and wide popularity of SIFT image features and bag-of-visual-words (BoVW) image representation, object retrieval has progressed significantly in the past years and has already found deployment in real-life applications and products, but the traditional BoVW methods have several problems, such as:low time efficiency and large memory consumption, the synonymy and polysemy of visual words. In this article, a method based on E2LSH (Exact Euclidean Locality Sensitive Hashing) and query expansion is proposed. Firstly, E2LSH is used to hash local features of training dataset, and a group of scalable random visual vocabularies are constructed. Then, the visual vocabulary histograms and index files are created according to these random vocabularies. Finally, a query expansion strategy is used to accomplish object retrieval. Experimental results show that the accuracy of the novel method is substantially improved compared to the traditional methods, and it adapts large scale datasets well.