In order to effectively utilize Web images to construct instructional resource database, a novel approach is proposed in this paper. With this approach, Web images and their semantics can be automatically downloaded, extracted and stored in resource database and the semantics can be refined by user feedback in retrieval progress. Image topic dictionary is built as the basis to extract semantics. Eight kinds of text are extracted as semantic source from Web pages. Based on image topic dictionary, image semantics can be extracted from the eight kinds of text. In order to further improve the accuracy of semantic extraction, we propose relevance feedback mechanism. Users can provide feedback to refine semantic annotation. The experimental results show that the approach is effective, in which high construction efficiency and quality can be achieved. The approach is better than manual annotation in efficiency and better than automatic annotation in accuracy. The similar methods can be applied to construct resource database of other forms of multimedia.