A Novel Multimedia Database System for Efficient Image/Video Retrieval Based on Hybrid-Tree Structure

With recent advances in computer vision, image processing and analysis, a retrieval process based on visual content has became a key component in achieving high efficiency image query for large multimedia databases. In this paper, we propose and develop a novel indexing mechanism built on the top of a Xindice database targeted at supporting quick access to image/video data along with strong retrieval functionalities for those high-dimensional metadata. The proposed database applies a native XML database as a backbone, uses hybrid-tree as its database structure, and performs threshold algorithm (TA) to conduct similarity search based on a random combination of image features. The database also sketches an index scheme in building a mapping between temporal and spatial domains so as to support both temporal and spatial queries at a higher speed search over the traditional pure temporal or spatial based indexing methods. The simulation results demonstrate that the proposed database system can conduct a more efficiently search activities supporting queries that involve composite features with arbitrary Boolean combinations and can outperform a traditional database employing a linear scanning search scheme