A video semantic object extraction method based on motion feature and visual attention

Abstract

In order to solve the semantic object extraction problem under complex background in video retrieval, a semantic object extraction method based on combination of motion features and visual attention is presented. First, moving object is extracted using motion features; then static object that has important semantic characteristic is extracted using visual attention; finally, a complete integration method is proposed to combine moving semantic object and static semantic object in the above. In this method, we take full advantages of motion feature of moving object and color and edge feature of static object. Experimental results show that this method has achieved remarkable results.