Research On High-dimensional Indexing In Content-Based Image Retrieval

With the rapid development of the Internet and multimedia technology, multimedia information increased dramatically, the number of images also grows rapidly. How to quickly retrieve the required information from the mass of the images is a important issue many applications are facing currently. An effective high-dimensional index mechanism is the key technologies of the search for large-scale images to achieve real-time requirements, is becoming more based on the content of the image retrieval in the field of research focus.For the current status and future development trend of high-dimensional indexing. Approximate retrieval methods overcome curse of dimensionality better. We study approximate retrieval methods and presents a similarity searches algorithm based pseudo grid filter and Local sensitive hash algorithm. The similarity searches algorithm is dynamic index algorithm. Firstly, it clusters the data objects, every cluster is small enough to be loaded into main memory, and then for each cluster using Local sensitive hash algorithm construct hash table and save hash table to the disk. Finally, using based on pseudo grid filtering algorithms to guide the hash table in the disk, to achieve good double filtering effect.Last, designing and realizing Da Meng Image Retrieval System-DMIR. The system is a distributed, multi-user image retrieval system, and supports color, shape, texture, and other query methods, has a good human-computer interface. In the system, the users can specify their own feature extraction algorithms and high dimensional index algorithms, and can manage and maintain algorithm, it has provided a experiments platform for high dimensional index research. For index part of the system, we apply the similarity searches algorithm based pseudo grid filter and Local sensitive hash algorithm, having achieved very good results.