Abstract
Search, retrieval and storage of video content over the Internet and online repositories can be efficiently improved using compact summarizations of this content. Robust and perceptual fingerprinting codes, extracted from local video features, are astutely used for identification and authentication purposes. Unlike existing fingerprinting schemes, this paper proposes a robust and perceptual fingerprinting solution that serves both video content identification and authentication. While content identification is served by the robustness of the proposed fingerprinting codes to content alterations and geometric attacks, their sensitivity to malicious attacks makes them fit for forgery detection and authentication. This dual usage is facilitated by a new concept of sequence normalization based on the circular shift properties of the discrete cosine and sine transforms (DCT and DST). Sequences of local features are normalized by estimating the circular shift required to align each of these sequences to a reference sequence. The fingerprinting codes, consisting of normalizing shifts, are properly modeled using information-theoretic concepts. Security, robustness and sensitivity analysis of the proposed scheme is provided in terms of the security of the secret keys used during the proposed normalization stage. The computational efficiency of the proposed scheme makes it appropriate for large scale and online deployment. Finally, the robustness (identification-based) and sensitivity (authentication-based) of the proposed fingerprinting codes to content alterations and geometric attacks is evaluated over a large set of video sequences where they outperform existing DCT-based codes in terms of robustness, discriminability and sensitivity to moderate and large size intentional alterations.