Robust Perceptual Image Hashing Based Temporal Synchronization for Watermarked H.264 Frames

Most H.264 watermarking schemes only embed the watermarks on I-frames, and are therefore extremely fragile under temporal attacks (e.g., frame dropping). To obtain the robustness under temporal attacks for H.264 watermarking systems, this paper presents a robust perceptual image hashing based temporal synchronization method. We extract robust geometric information from the 4times4 block mean map of the luminance values using a Singular Value Decomposition (SVD) based image hash algorithm. By searching frames with the minimum hash distances from the original watermarked I-frames, the video at the detector is synchronized with the watermark signal, which thus avoids invalid detections on unwatermarked frames. Experimental results demonstrate that our synchronization method can maintain reliable detection performances on the watermarked video with dropping frames even when common spatial attacks like Gaussian noises and recompressions also exist.