Massive WWW Image Retrieval Based On Automatic Annotation

As an important information carrier, image is an essential part of the network of information resources. With the rapid development of digital technology and scanning technology, we can access increasing image data on the internet. How to identify the image information in needs more quickly and accurately from vast network image data is the current hotspot in image retrieval area. Therefore, on the basis of the former search about search engine and annotation based on image content, this paper puts forward a massive WWW image retrieval framework based on automatic annotation.The massive data index of image retrieval can not be finished in concentrically retrieval system independently, and the distributed system is a basic approach to resolve the problem of data size and system scalability. So the massive WWW image retrieval framework is based on distributed system. This paper introduces the design and principle of a distributed retrieval system. It describes key technologies and implements of the important modules in the system development. To take advantage of distributed system and improve system performance, the goal of our implements is the system better scalability, adaptability and more fault-tolerance.Retrieval framework query is based on keywords that the user inputs. For the network images, the text in its website contains rich semantic information related to the images. The retrieval framework starts with extraction of semantic keywords from the related text. This step is similar to ordinary text retrieval. However, sometimes the website text is insufficient to provide semantic keywords. So we propose a new approach of image annotation based on image content. When there is no effective related text for the images, the system annotates the image with the approach based on machine learning and statistical model automatically. The annotation will be used as the keywords for the image retrieval. Our approach is based Co-occurence model. We enhance automatic annotation effect in 3 ways: 1) Effective image segmentation and image object extraction. 2) Input images will be mapped to more than one cluster, thus obtaining more precise semantic annotation. 3) Heavier the importance of the objects in the image center. Experiments show that the technology can effectively improve the recall and precision in network image retrieval system.