A new framework of CBIR based on KDD

The research emphasis of CBIR (content-based image retrieval) was put on the low-level visual feature extraction to resolve the problem of manual annotation for images in recent years. But because of variety of images, the extracted visual features can not express the semantic content of each image correctly. So the retrieval accuracy is lower than the text-based image retrieval accuracy generally. In this paper we propose a new retrieval method based on KDD (knowledge discovery in databases) and knowledge reasoning to improve the retrieval accuracy and relate the low-level image features with the high-level semantic content. Experiments show that the result is much better than the traditional retrieval method.