Perception-driven Watermarking with Evolutionary Block Mapping

This paper presents a perception-driven, block-based wavelet watermarking algorithm, which merits in both perceptual invisibility and robustness against watermark attacking. First, attention-based region saliency analysis is adopted to select perceptually unconcerned regions for watermark embedding. Second, wavelet transform is carried out to embed secret message into high frequency coefficients within the perceptually unconcerned regions. Especially, we present an optimal block mapping substitution method, which adopts genetic algorithm to reduce PSNR cost. Its main idea is to minimize the degradation of the watermarked image by learning a best mapping function between original image blocks and secret messages at global scope. Extensive experiments are conducted to validate both the perceptual invisibility and robustness against Gaussian Noise and JPEG compression attacks. We also give PSNR comparison with state-of-the-art methods to demonstrate our advantages.