Identification of Trademarks Painted on Ground and Billboards Using H.264 Compressed Domain Features from Sports Videos

摘要

In this paper a novel methodology to automatically identify and localize the Trademarks from video using compress domain features of H.264 video is proposed. We have used the H.264 video stream as input and use some compress domain features obtained during parsing and decoding process of the stream to identify the text regions. The features used in our approach are (i) Edge strength information for each Macro block obtained from de-blocking filter edges and (ii) DC component of integer transformed luma coefficients. These features are used to localize the text regions. Also authors had proposed a decision making process based on these features to determine whether a recognized region is score-card or a trademark. Experimental results shows that the proposed approach can work in real time in any commercially available DSP platform and can detect the trademarks successfully. The system on different type of sports videos gives a recall rate of .95 and a precision rate of .92