Research On Clustering-Based Indexing Approach In Image Retrieval

With the development of multimedia and Internet technique, a number of image information appears. Traditional Text-Based Image Retrieval (TBIR) cannot meet the need of image information retrieval. So more researchers, recently, began to focus on Content-Based Image Retrieval (CBIR).At first, the background, development and application of CBIR for universal techniques are summarized. This paper presented the development of CBIR technology, the key technology of CBIR and evaluation of CBIR. To find out the merit and shortcoming of various calculate way, we analyzed and compared some multi-media dates with the comparability searching calculate way. It provided theories foundation for bringing forward a new mode or a new calculate way of image retrieval. In 30 years to research on the data index with high-dimension vector space, there was already a great deal of index structure. We not only introduced and classified the existing high-dimension index in the thesis, but also introduced wavelet analysis technology, clustering technology, index technique and so on.Author introduces a new pattern efficient, dynamic clustering-based multi-layer index structure. This algorithm uses the multi-scale wavelet technology to process the distribution curve of samples, and uses multi-layer dynamic clustering to create a multi-layer index structure. Meantime the system automatically storage the information of clustering into the database. This index structure effectually supports data retrieval, dynamic inserting and dynamic deleting, and evidently reduces the covering with dates, guarantees updating the index structure in time.The theoretical basis of the above, the author designs a CBIR experimental platform. Compared by inserting, deleting and retrieval efficiency and analyzing some special cases, it is convincingly proved that hierarchical index structure based on dynamic clustering is efficient in CBIR.