Research On High-dimensional Index Technique And Its Application On Medical Image Database System

Recently, increasing researchers focus on high-dimensional index technology, which is a science of improving the retrieval effectiveness of high-dimensional database by establishment of index structure. As an important part of high dimensional database, image database's retrieval ability relies on the support of high-dimensional index technique.Existing high-dimensional index techniques are compared and analyzed. By systematically analyzing of relevant algorithms, high-dimensional index techniques are investigated from theory, algorithm and implementation aspects, and a novel method of index is presented. In additional, a prototype retrieval system of medical image database is designed and implemented based on this novel method. The main work is listed here:1 , By study high-dimensional data space searching, high-dimensional data and its index technique are particularly analyzed. The relevant aspects of high-dimensional data query are discussed. The basic idea, structure and algorithms of high-dimensional index are concluded.2, Indexing idea , structure and applicability are compared between metric access method (MAM) and spatial access method(SAM). The similarity and difference of those two methods are summarized. Applying each method to image database retrieval, we find study fields of image database index technique.3, A new index method of M*-tree are proposed. On the basis of M-tree's structure, this method, called as M*-tree, makes use of a global multi-way inserting algorithm for dynamic building index tree and exploits an ameliorated slim-down algorithm which can optimize its structure. M*-tree improves nodes utility of index structure, decreases the quantity of overlaps between two regions of nodes and has a prominent advantage that it can be applied to high or hyper-dimensional data retrieval which is proved by experiment.4, On the basis of M*-tree a retrieval system of medical image database is designed and implemented. According to the characteristic of medical image information, this system takes use of the method of maximum entropy threshold segmentation to extract regional metric histogram as features and applies M*-tree method to index and organize of features database. Extracting feature is simply and the effect of feature index is obvious in this prototype system that improves performance of image retrieval effectively.