Robustness Evaluation of Perceptual Watermarks

In a watermarking system, robustness evaluation should be made if invisibility criteria are satisfied. For this purpose, we propose to measure the perceptual impact of a watermark on the image using the well known peak signal-to-noise ratio (PSNR) and additionally a spatial masking model that indicates "bad" pixels between the original and watermarked images. Two perceptual watermarks using pixel-wise masking in the wavelet domain are being tested, limiting the watermark strength such that the PSNR is 35 dB and in average the percentage of affected pixels is less than 25%. Extensive tests were made on several watermarked images and for different types of attacks (JPEG compression, median filtering, resizing, cropping and gamma correction) in order to see the effect of these attacks on the detector results. Simulations show some limitations of these techniques under the invisibility constraint.