Abstract: Multi-concept learning is a very important technology of video retrieval. This paper gave a method of multi concept learning based on inter-concept relation. First, analysed the types of inter-concept relation in large-scale multi media lexicon, then got ctuantitative analysis with TRECV)D2005 annotation data and Columbia374 data. Second, selected correlative concepts to build context based conceptual fusion detectors, then got weights with reliability of detectors.At last, fused the two detectors with visually similarity. The results show that the proposed method achieves more remarkable and consistent improvement.

Key words: Multi-concept learning, Inter-concept relation, Concept detectors