Fusing audio-visual fingerprint to detect TV commercial advertisement

Abstract

Sixty-four percent of consumers believe television advertising still has the greatest impact on them. Therefore, there is a great application to provide accurate and real-time TV advertising identification for government and advertisement providers. As the integration of multi-modal method takes full account of video and audio information, this paper aims to handle composite fingerprinting in a unified framework for advertising identification. The Improved Harris Combining Motion feature which is based on the differences between the adjacent video frames can produce video fingerprint. Meanwhile the proposed FIR filter based Fast Audio Fingerprint is focused on extracting the differences between the equivalent bands from adjacent frames. Moreover, this multi-model framework combines the audio and video fingerprint by weighted manner. Experimental results show that compared with the current methods, both audio and video fingerprint has the advantage of higher discrimination, stronger robustness and lower time complexity. Moreover, multi model fingerprint can enhances the performance of the unique fingerprint.