Research Of Image Retriecal Methods Based On Internet

With the rapid development of the multimedia technology and internet technology, information on the internet contains not only simple text data, but also large amounts of audio, image, video and other kinds of multimedia information. How to effectively manage a large number of image information, and that from which the images can be retrieved efficiently has become an urgent problem. In order to achieve fast and accurate image retrieval, Web-based image retrieval techniques have been developed and become one of the hot areas of research on the image analysis.There are already many search engines to provide image search services such as Google, Baidu, Soso, Ditto, PicSearch, Ixquick, Mamma, etc. but these search engines are all based on keywords or descriptions texts, so they are essentially the text-based image retrieval. This type of search technology will eventually be converted into a kind of the database search aiming at the text, featuring a retrieval theory being already very mature. Nonetheless, there may be a fatal problem, that is, text labels are required on the image, and the workload is huge. Apart from that, the "semantic gap" is another problem, that is, low-level content feature of an image is unable to provide an effective description of the high-level semantics.Based on the analysis of the research survey of home and abroad, combined with the existing research work, this paper mainly provided the research results on the Web-based image retrieval technique with regard to its several key issues, where four aspects are discussed on common visual features, including search technology, feature selection, dimensionality reduction methods, and similarity measure. Major research work and innovations of this paper are as follows:(1) Research and analysis are given to the issues that may be found in image retrieval techniques, followed by introducting the technology of image retrieval based on visual feature, paving the way for the follow-up study on Web image retrieval.(2) Targeting at the traditional genetic algorithm for solving the image feature selection optimization with the problems that may be premature or lack in local convergence, the adaptive genetic algorithm is combined with the principle of parallel computing applied in genetic algorithms. Using the improved mutation operator and crossover, a genetic algorithm based on two-population adaptive random probability operator is proposed and successfully used for image feature selection and optimization, achieving better performance.(3) A dimension reduction method based on the HSV color model to extract dominant color is proposed. RGB color model is converted to the HSV space based on visual perception. Then, the image is undergone with 72-dimensional vector of HSV features. This is followed by sorting in a descending order by the feature value of each dimension. According to the indicated threshold, the first d dimensions are taken as the number of the intrinsic dimensions of an image, so as to achieve the purpose of quick dimensionality reduction. Studies have shown that this dimensionality reduction method can be applied without specifying the dimension number of the required reduction, and it can take advantage of the intrinsic dimensions of an image to achieve the purpose of dimensionality reduction.(4) An image measurement method is proposed based on the area similarity of mutual information. On the basis of the Shannon entropy, joint entropy, conditional entropy, Shannon mutual information between images is taken as a similarity functions. For the issue on Shannon mutual information that requires three normalizations involving a large amount of computation, Shannon mutual information calculation is optimized to reduce the computational workload by the image space, thus improving the image retrieval performance.(5) Research is performed on the model framework for Web-oriented image retrieval system, initially building up an image retrieval system based on image content . Specifically, the system model is divided into interface design, web spider, image pre-processing, image feature extraction and selection, similarity measure and the feedback. Image search function is added up to the text search engine, creating 72-tree for image index and classification, achieving really Web image retrieval based on target image, by achieving a Web page image capture, database storage and similarity measure.