EMD similarity measure and Metric Access Method using EMD lower bound

Abstract

In this paper, we propose a method of indexing the color images, and an index structure of a large image database. The first method combines color’s information of the pixels and their positions in the images. It also aims at reducing the size of signatures of each image, and consequently reducing the computation time between two signatures. The technique used for this purpose is the Kd-tree. The second contribution proposed in this paper is the LBMTree (Lower Bound MTree) structure, which is a method access metric. The LBMTree is an extension of the MTree structure. This structure uses the lower bounds of the EMD metric. Our goal is to improve the search time and maintain the effectiveness of the EMD metric in terms of accuracy.

Experiments show the performance of research in the LBMTree compared with those obtained by the sequential scan, and the research in the MTree. In the other hand the performances of the proposed signature and of the EMD metric are measured, and results are shown.