The Research Of Image Retrieval Technology Based On High-Level Semantic

Along with the multimedia technical violent development and the Internet technology increasingly popularization, the origin of digital images are more and more expanding, the type and quantity of images data are also growing day by day. Regardless of military or civil equipment, they can produce the capacity of equal to several kilo mega bytes of images every day, and the retrieval of image information is a challenge to traditional retrieval methods. The image retrieval system based on low-level feature can't satisfy user's retrieval needs. In this thesis proposed semantic-based image retrieval method is an effective to image retrieval. And the research aims at image semantic classification and retrieval.In this thesis, we firstly review the content and the state of the semantic-based image retrieval research, especially including the contact between image low-level features and high-level semantics, and some typical approaches to extract image semantic are presented; secondly, an image retrieval method based on color edge gridding histogram is proposed, the method uses the important edge information and considering the spatial distribution of color information to image retrieval; According to the idea of semantic image retrieval, we provide a method based on Support Vector Machine (SVM) to map image low-level features to high-level semantic, which owes to the superior performance of SVM on statistic. We can express image semantic extracted by SVM as image high-level content, and color or texture feature as image low-level content, and then we can retrieval images by these two expressions. In discussion, we focused on the experiment on an especial image database to realize image retrieval and semantic annotation, and our experimental results show that the mapping method is effective in expressing the ordinary user's semantics.