The Research Of Content-based Image Indexing And Browsing Algorithm

With the rapid development of Internet and multimedia technology, the amount of digital information is increasing rapidly. Therefore, research on image retrieval is attracting more and more attention. Content-based image retrieval perform retrieval according to the visual feature of images, can support image retrieval efficiently, it has become an important topic in multimedia retrieval fields. High-dimensional data space indexing scheme, organizing and browsing of the retrieval results have become the hotspot in content-based image retrieval fields.Efficient high-dimensional indexing scheme is required for real time retrieval in large-scale image database. Many index structures have been proposed to solve this difficult problem, but as the dimensionality increasing, the query performance of most index structures will suffer greatly. By systematically analyzing existing high-dimensional indexing techniques and relevant algorithms, we make full use of the statistic characteristic of the first principal component to prune data space, and propose a novel index structure , called P2-tree. Firstly, we apply principal component analysis on original data space ,then use top-down clustering scheme to recursively decompose the dataset, combine triangle inequation property and principal component to prune data space when query. The experiment results show that the proposed index structure based on principal component has good efficiency for reducing the times of distance calculation, can support similarity query effectively.Most of existing image search engines are text-based retrieval model, and image search results browsing is ranking-based list presentation. Because this model ignores visual-feature of image, it is time-consuming process for users to find images of interest in returned image collection. This paper explores the problem of browsing Web image search results, applies information result visualization technique, proposes a novel image browsing algorithm based on nearest neighbor visualization .The algorithm combines methods of semantic-based and visual-based retrieval, analyses retrieval results of image search engine and organizes the results based on visual similarity. A nearest neighbor search algorithm based on keyindex is proposed in this paper, which can be used to improve the performance of similarity search. Experimental results show the validity of algorithms, it can enhance themutuality of image retrieval system and user, can help user to explore image search results more naturally and efficiently.