A Strategy of Semantic Information Extraction for Web Image

In order to improve semantic information extraction coverage and quality of Web images, A novel approach is presented in this paper. Based on the habit of users to retrieve images, the model of representing image semantics is put forward. According to the model, a series of dictionaries are built, which include image topic, object and attribute dictionaries. Eight kinds of text are extracted as image semantic source from Web pages. Combining with semantic dictionaries, image semantic keywords can be extracted from the eight kinds of text. The strategy of extracting image semantics is better than existing technique, which is better than manual annotation in efficiency and better than automatic annotation based on content in accuracy. A performance experiment is presented which shows that high extraction coverage and quality can be achieved with this approach. The similar approaches could be applied to extract semantic information of other forms of multimedia.