Improving search time for contentment based image retrieval via , LSH , MTRee , and EMD bounds 1

Comparison of images requires a distance metric that is sensitive to the spatial location of objects and features. The Earth Mover’s Distance was introduced in Computer Vision to better approach human perceptual similarities. Its computation, however, is too complex for usage in interactive multimedia database scenarios. Nearest neighbor (NN) search in high dimensional space is an important problem in many applications, in particular if the method of similarity measure used be the EMD. Ideally, a practical solution (i) should be implementable in a relational database, and (ii) its query cost should grow sub-linearly with the dataset size, regardless of the data and query distributions. Despite the bulk of NN literature, no solution fulfills both requirements, except locality sensitive hashing (LSH). In this paper, we propose a new index structure, named LSH-LUBMTree, for efficient retrieval of multimedia objects. It combines the advantages of LSH [27,28], the technique used to embedding [33] the EMD, and the advantages, of LUBMTree [19]. Unlike the images of each bucket are stored in the LUBMTree. Experimental results show that LSH-LUBMTree performs better than the standard LSH in term of search time.