Indexing Technology In Content-Based Image Retrieval Under A Web Environment

Recently, advances in Internet and World Wide Web technology have led to exponential proliferation of image information, so that Content Based Image Retrieval (CBIR) technology has been widely studied.In general, CBIR builds a feature database of image by abstracting the features of images, in the proceeding a single image is turned into a point in high dimension, which is called a feature vector. In this way, similarity-searching problem in high dimensions is turned to nearest-neighbor (NN) query problem. One of the most important issues about it is how to build index structure.In many cases it is not necessary to insist on the exact answer; instead, determining an approximate answer should suffice. Following this direction, the famous Locality-Sensitive Hashing (LSH) has been introduced. The LSH like approach enabled us to achieve fast approximate nearest neighbor with a bounded error rate in sense of statistic. The PLSH (PDF LSH) index structure is an extended version of original LSH.PLSH support the approximate NN search under the PDF measure, it improves the accurate of query while not losing efficiency, the correctness of PLSH can be proved.Besides, the parameter auto tuning method of PLSH enable index structure to fit to different dataset. Further more, PLSH structure preserves the interface of parallel processing.Experimental results indicate that PLSH has better accurate than original LSH, while the taking almost the same time.