An efficient method to recognize caption area in MPEG compressed video was presented, by making use of the contrast of I|frame to distinguish caption area with background. We define texture energy, intensity of boundary, distance of background, and texture correlation to recognize caption area and caption frame. The benefit of only analyzing I|frame is that we can make use of DCT coefficients directly without losing information. We have experimented with our algorithm, and the result of experiment indicates that the performance of the algorithm is efficient.
展开▼