Video Watermarking Based on 3D Scale Invariant Spatio-Temporal Feature Points

Abstract

3D SIFP (3 dimensional scale invariant features points) can embody the great changes in video both in spatial and temporal domain, which can assure the stability against spatial and temporal attacks. As a result, they enforce the robustness of video watermarking. In this paper, a novel video watermarking scheme exploiting 3D SIFP in DCT (Discrete Cosine Transform) domain is proposed. We establish the 3D difference of Gaussian pyramid (3D DoGP) and 3D Hessian matrix to locate the 3D SIFP, which are selected by global and local steps. All global SIFP are ordered in ascending order according to their 3D Hessian matrix response values, and SIFP corresponding to the first several global response values are selected to maintain the stability. Afterwards, the SIFP with the largest local response value in detected frame are selected as the center to generate the square embedding region. The region is transformed into DCT domain, and the ZigZag scanned mid-frequency coefficients are segmented to embed the watermark by using modified odd-even quantization. Experimental results show that the proposed scheme guarantees high peak signal to noise ratio (PSNR), and is very robust against noising, filtering, JPEG compression, frame swapping, frame insert, frame dropping, scaling etc.

Keywords

• Video watermarking
• Scale invariant
• Feature point
• Robust regions
• SIFP.