The Key Technology Research On Semantic-based Image Retrieval

Semantic-based image retrieval technology is the future development direction of imageretrieval technology; it can be effectively applied to all kinds of professional image retrievaland the image queries of Internet. Semantic-based image retrieval can divide images intocategories according to the understanding of people. And then is combined with semanticcategories and bottom image characteristics to achieve image retrieval. Penple can use thesemantic-based image retrieval technology better access to information, thus to perceive theworld. The image semantic multi-class classification is a key technology of semantic-basedimage retrieval. In order to retrieve the corresponding picture information more quickly,people often use hierarchical semantic to express images. This method can retrieve imagefrom different levels. Currently, Support Vector Machine (SVM) is commonly used to solveimage classification. However, with the increase in the amount of picture data and in thegreater varieties, massive information of pictures produces a lot of tough challenges to thesemantic image classification technology. In order to improve the efficiency and accuracy ofthe image classification, Fuzzy Support Vector Machine (FSVM) theory is creativelyintroduced into the image classification in this paper to solve the contradiction between theefficiency and accuracy of the image classification for a particular purpose.This paper introduces the theory of Fuzzy Support Vector Machine(FSVM) and makesimprovement on FSVM which eliminates the Support Vector Machine(SVM) classifier’sunclassifiable region so that it enhances the image classification accuracy rate of lower-levelsemantic image, and laids on the basis for the high-level semantic classification.This paperestablishes the mapping between bottom image characteristics and low-level semantic image,and realize the link of low-level semantic image and high-level semantic image. It breaks thetraditional description approach of hierarchy semantic, and achieves the hierarchical semanticdescription structures. The experimental results show that the method presented can improvethe image multi-class classification accuracy rate, especially the image classification accuracy rate of high-level hierarchical semantic image. This paper results can be applied to imageretrieval for increasing the rate of recall and precision rate.