Abstract: For bridging semantic gap between video low-level features and high-level semantic concepts in the semantic-based video retrieval system,the video multi-semantic annotation algorithm based on feedback fuzzy graph theory was proposed．First,a training set which includes most temporal and spatial distribution of the whole data is made up and it will achieve a satisfying performance even in the case of limited size of training set．Secondly,the fuzzy operators are applied to graph theory to achieve fuzzy reasoning by using fuzzy semantic．Last,in order to finish the feedback of video annotation,some temples from the testing set that have finished annotation are selected and added into the training set．Experimental results indicate that feedback fuzzy graph not only sets up the relationship between semantic concepts well,but also improves the precision of annotation and shows good performance.

Key words: Video annotation,Fuzzy graph,Multi-semantic annotation,Semantic gap