| Recently, with the rapid development of multimedia technology and computer networks, as the important part of multimedia data, the capacity of digital images is growing at a sharp rate. How to retrieve required information from tremendous image data efficiently and rapidly has become a pressing issue.The traditional image retrieval techniques, which are based on text description and keyword, are no longer meeting the demands of users. At present, Content-Based Image Retrival (CBIR) technology is adopted generally. The main idea of CBIR is to extract the feature vector of images according to the color, texture, shape and etc from images, and then store the feature vectors in file or database, afterwards, calculate the similarities between query images and images in the database, and then return the results.As the capacity of the image data is tremendous, and there are many features to describe an image. So, the feature vectors of images are usually high dimensions. Therefore, the retrieval relies on the support of high-dimensional index technique, which is a technique to improve the retrieval effectiveness of high-dimensional database by establishing index structure.Lots of researches to index multi-dimension data have been done effectively, and many index structures, such as R-Tree, KD-Tree and SR-Tree, have been proposed. Their performance decreases sharply with the dimension increasing, although they work well in low dimension space. This phenomenon is called "the curse of dimensionality".CBIR technology is introduced and existing high-dimensional index techniques are compared and analyzed. By systematically analyzing of relevant algorithms, high-dimensional index techniques, a novel method of index considering the angle information on the base of iDistance structure, called A-iDistance, is presented. At the end, the experiment results on medical images and synthetical data show that the proposed index structure works well in high dimension space. |
PDF Full Text Request