University of Central Florida at TRECVID 2007 Semantic Video Classification and Automatic Search.

摘要: In this paper, we describe our approaches and experiments in semantic video classiflcation (high-level features extraction) fully automatic topic search tasks of TRECVID 2007. We designed a unifled high-level extraction framework. Two types discriminative low level features, Spatial Pyramid Edge/Color Histograms Bag Visterms, are extracted from the key-frames shots. Then SVM classiflers with RBF kernel used for classiflcation. The flnal results produced by fusing combining these classiflers. experiment show that combined substantially improved performance over individual feature based classifler. task, mostly focus on retrieval using visual content through high detectors. main challenge task is mapping queries to features. A novel earth mover’s distance (EMD) relevance procedure flnds similarity between videos word measures.