TV logo recognition based on luminance variance

As the rapid development of intelligent TV, TV logo recognition plays an important role in the applications of video analysis and video retrieval. This paper utilizes the luminance variance information to extract logo mask along a set of consecutive frames. An effect and efficient method, projection distribution histogram intersection (PDHI) is proposed to calculate the similarity of two logos. Experiments are conducted on a large database collected by ourselves, including 57 challenging standard definition (SD) and 10 high definition (HD) TV channels, the results are promising.