The dissertation focuses on key techniques in web image retrieval. They are image auto-annotation, relevance feedback in searching process, and web image mining semantically. The main contributions of the dissertation are as follows: (1) The discussion and analysis about image auto-annotation are given in details. Based on some related work, we propose a unified annotation framework via multi-graph learning, which includes two sub-processes, i.e., basic image annotation and annotation refinement. In the basic annotation process, image-based graph learning is utilized to obtain the candidate annotations. In the annotation refinement process, the word-based graph learning is used to refine those candidate annotations from the prior process. (2) Under the direction of the proposed annotation framework, we propose some effective approaches to estimate the multi-graph model. Specially, NSC-based method is used to construct the image-based graph model, statistical information and search-based approaches are utilized to construct the word-based graph model. The two sub-processes are performed sequentially and finish the task of image annotation effectively. (3) A dual cross-media relevance model (DCMRM) is proposed for automatic image annotation, which provides a new direction to the study of image auto-annotation. To the best of our knowledge, we are the first to formally integrate image retrieval, and web search techniques together to solve the image annotation problem. This relieves the dependence on training set and makes the large-scale image annotation possible. (4) A human behavior consistent relevance feedback model for image retrieval is designed. Simulating human behaviors, the proposed model enable the user to perform relevance feedback in three manners: Follow up, Go back, and Restart. Each manner is a way for the user to provide the system with his or her opinions about search results. The accumulated feedback information can be used to refine the user query and regulate the similarity metric. We adopt the graph ranking algorithm to model the retrieval process. (5) A method of web image mining semantically with the help of a web search engine is proposed in the dissertation. The method can be automatically performed to mine many web images relevant to a specific concept. By repeating the automatic process many times with different concepts, a large scale image set can be obtained easily. That is, the method has good scalability.
修改评论