A classification-matching combination for image retrieval

Nowadays, applications dealing with information extracted from images are c ommonplace. The widespread use of multimedia information (images, video, au dio etc.) makes necessary applications capable of storing, and therefore re trieving, it. Information extracted from images is usually complex and high dimensional. The extraction of non-textual low-level indexing features fro m images is now a research field, and this process principally suffers beca use of the computational cost of the high dimensionality of those features. A new way to classify and match low-level features extracted from images, for retrieval purposes, is presented in this paper. M-tree and R-tree struc tures are used, as well as an incremental version of the k-means classifica tion algorithm. This set of algorithms is used to solve the problem of low performance when retrieving previously catalogued images.