Video fingerprinting by using boosted features

Abstract

In this paper, we present a novel approach for video fingerprinting by using boosted Harr-like features and direct hashing. Through employing a pairwise boosting method on a large set of features, our system can learn the top-M discriminative filters that are enable to efficient extracting video fingerprints. During query phase, we retrieve video clips by using a fast and accurate direct hashing, which minimizes perceptual Hamming distance between queries and a large database of pre-computed fingerprints. To demonstrate the superiority of our method, we also implement four other fingerprinting methods for comparisons. The experimental results indicate that our proposed method can significantly outperform those four methods in video retrieval. © 2009 Springer Berlin Heidelberg.