Research On Semantic-Based Image Annotation And Retrieval

With the rapidly developing of multimedia and network technology, the image data is expanding. People are facing with a difficult problem that how to effectively use a large number of images. It has always been a hot issue that how to use image resources through the image annotation and retrieval. In this paper, some key problems are researched in this thesis, and some results are obtained:Proposed of the image shape semantic features. In this paper, some appropriate underlying features are selected to generate the shape semantic feature. Adding shape semantic features generated from shape features to enhance the semantic information. Experiments show the effectiveness and retrieval robustness of shape semantic features.An approach to content-semantic-based image retrieval is proposed. The shape semantic features are used in image retrieval. And the content-based image retrieval and semantic-based image retrieval is combined together. It makes the retrieval direction correct and also improve details of the retrieval result.VAWO-based and word correlation-based image annotation method is proposed. Extracting the most interesting part from the image, and also consider the focus area of the people in the visual understanding of the image. The inner relection of the various objects in the image is calculated by the wordnet. Through experiments show that the VAWO and word correlation-based image annotation improves the accuracy of image annotation.A system of image annotation based on VAWO image and word correlation is implemented. And a system of image retrieval based on content and semantic is implemented. Image retrieval system supports text query and image query, and also supports sorting of image retrieval results.In summary, this paper proposes four methods to improve image annotation and retrieval. And a system of image annotation and retrieval is developed. Finally, the experiments result presents the significance and worthiness of the research.