Research On Technologies Of Image Semantic Annotation And Retrieval

With the development of computer and network technology, the research focuses how to effectively organize, manage and fully use multimedia information, how to fast and effectively query and retrieval needful them in current domain of media information process and understanding. For unstructured image data, the traditional textual retrieval method and the Content-Based Image Retrieval (CBIR) can not solve the problems in semantic based image retrieval. In the practical application, users need to query images based on the high level concept features. In order to bridge the"semantic gap"between visual features and concept features, this paper designs and implements a prototype system of image annotation and retrieval based on semantic. Its novelty includes the following three aspects:1. Construct image semantic knowledge for special domain and new MPEG-7 description scheme.On the level of the objects and their relationships in the image, an image semantic knowledge base for special domain is constructed by adopting the feature of ontology specification to describe domain knowledge. A new description scheme is acquired by extending MPEG-7 multimedia description scheme, providing a new description mechanism that can be conversed to the description in MPEG-7.2. Acquire and annotate of image semanticIn the paper, image segment algorithm is adopted to acquire its objects, and regional integration technology is utilized to acquire the final image semantic objects with user's interaction. According to the semantic ontology knowledge, MPEG-7 and graphical method are adopted to organize image semantic content, which is the final format of user's understanding.3. Retrieve images based on semanticAccording to the graphical form of semantic, the indexing mechanism based on path is constructed to implement the image retrieval based on semantic. In the implementation of retrieval system, Lucene search engine is adopted to construct the index, and Xpath is utilized to process XML documents.The corresponding experimental data is compared with the method of full text on recall and precision. The results testify that our method can effectively implement semantic based image retrieval.