Image Retrieval Based On Semantic Learning

Recently, the techniques of content-based image retrieval (CBIR) have been achieved great developments. The most existing systems of CBIR fulfill the tasks of retrieving the similar images through computing the degree of similarity of different images. Though those methods have achieved much success, they all have great limitations. The main difficulty in the state of art is that, the qualities of retrieval results by computers are much dependent on the features of images, and they have great differences with human beings who predict the degree of similarity of images through the image semantics. The differences make the retrieval results through most existing methods unsatisfied. To solve this problem, a novel method is proposed in this dissertation for learning the user semantics, which is got from several similar images selected by the users for retrieving the similar images in the database. To obtain the good features for learning image semantics, the methods of color quantization and image segmentation which is improved from JSEG are proposed. The JSEG algorithm only utilized the color distribution of the quantized image; while, the improved algorithm adds the analysis of color and texture information to improve the result of image segmentation. Because there is lack of researches on semantic concept saving after the good retrieval results which have the good semantic concept hidden in similar images, a method of semantic saving based on the research of complex networks is proposed to facilitate retrieving the future similar semantic images. The experimental results show that the retrieval results of learning user semantics method and of semantic concept saving method are both satisfying.According to the idea mentioned in the last paragraph, this dissertation firstly introduces the research of CBIR in whole, which is included such as the origin of CBIR, the development, the main problems confronted in CBIR, the research areas and innovations in the dissertation.