Matching Measures in the Context of CBIR: A Comparative Study in Terms of Effectiveness and Efficiency

摘要

In this paper, we compare between many matching measures (distances, quasi-distances, similarities and divergences), in the context of CBIR, in terms of effectiveness and efficiency. The major effort put up to now, within the area of CBIR, is usually made into the indexing stage. This work then highlights the importance of the matching process, as an important component within the CBIR system, through making under experimentation a large number of matching measures. The experiments, conducted on the Wang database (COREL1 K) and using two signatures: histograms and color moments, reveal that Euclidean distance, usually used in the context of CBIR, gives less performances comparing to the Ruzicka similarity, Manhattan distance and Neyman-X~2.