Based On The Research Of Image Automatic Annotation Of Image Retrieval System

Image is the most intuitive carrier for people understanding the world. In recent years with rapid development of the multimedia digital technology and the rapid popularization of the Internet, Immense amounts of image data flowed into the computer network and increased in a geometric ratio. However, such immense amounts of image data have not translated into information and knowledge. The use of the image resources was next to nothing.The reason for this was that it is difficult to quickly and accurately to find the information we need in immense amounts of image data, so-called image retrieval technology is still too young. From the 1970s there has Image Retrieval technology, mainly Text-based Image Retrieval which referred TBIR. TBIR is strict in early manual image annotation, and has serious limitations. Especially in recent years, the source of image is rich and uncertainty, which make the standard manual annotation become impossible. Since the 1990s there appeared the Content-based Image Retrieval. It has achieved search technology which based on color histograms and texture histograms in the low level. But people’s understanding and demand of image mainly in the image semantics, so this dissertation chooses the automatic image annotation system based on image semantic as the research subject. The main contributions of the dissertation are as follows:First, summarizes the development process of image retrieval. The image mining and automatic image annotation technology were studied.Second, designed a system based on classification. it consists of automatic image annotation and image retrieval.Third, we analysis the major problem in Image Near-duplicate Retrieval based on Bag of Words model as visual polysemy and synonymy phenomenon. To eliminate this phenomenon, we propose using associate rule mining to find visual pattern. We propose and compare different usages of visual pattern. Experiments prove our proposed method is superior to classic model.Four, presents an AS-BoW model to optimization of automatic image annotation retrieval results.