The Research On Image Retrieval Based On Content

With the widening usage of image, the technology of organization and management and retrieval of image source is becoming more and more important. In order to improve the utilization of image source, image retrieval from the huge image database with high speed and efficiency is needed to be solved urgently. The technology of content-based image retrieval(CBIR) and semantic based image retrieval are effective approaches to solve the problem.The paper introduces the background,the application and and some representative systems of CBIR, and expatiates the key technique of content-based image retrieval in detail. After researching relevance feedback, we firstly focus on the image content multiple interpretations and then do research around this subject. The image possibilitistic membership can express the multiple interpretations of an image. In light of the image possibilitistic membership, a new image retrieval method with relevance feedback based on possibilistic cluster is proposed in this paper. The method firstly classifys images in image database using the possibilistic cluster algorithms,then just inquiring images in the existent classification. The paper also proposes a new relevance feedback image retrieval algorithm, features in which the user is especially interested will be chosen as the attributes in image retrieval according to user's preference feedback.In object based image retrieval method, whether a picture is the user's requirement or not is determined by whether it contains a special object or not. Face object based image retrieval is valuable in application. Its key technique is face recognition. Face recognition means recognizing or verifying one face(or several faces) from the face image database using image processing and pattern recognition technique .So the paper also do some researches on the method of face detection and face recognition and the design classifing method. According to theadvantage in reducing training time and optimizing network topology architecture of rough neural network, a novel face recognition method is presented based on multi-features using fusion of multiple rough neural networks classifiers.