RST invariant video watermarking based on log-polar mapping and phase-only filtering

Abstract

In this paper, we present a video watermarking algorithm based on the log-polar mapping and phase-only filtering method. The log-polar mapping (LPM) domain is obtained first from the magnitude of the Fourier spectrum of the frame. Then, the watermark pattern is embedded in the LPM domain based on the feature that rotation and scaling transformations in the spatial domain result in cyclically translational shifts in the logpolar mapping domain. A matching template is cut from the LPM domain after watermark embedding and is used for correlation matching to find the RST parameters for the watermarked video undergone geometric attacks. Our new phase-only filtering method is used and it is the only filter that can provide an acceptable discrimination while rotation or scaling or both applied to the watermarked video. A square portion from I-frame of each Group of Picture (GOP) is used for watermark embedding based on our analysis that a square image has better tolerance to rotation attack than a rectangular image. The experimental results demonstrate that this algorithm is robust against rotation, scaling, translation transform, noise addition, filtering, MPEG-2 compression, etc.