Removing spatial spread spectrum watermarks by non-linear filtering

Many watermarking methods are based on adding pseudorandom noise in the spatial domain. In general, the robustness of the watermark is determined by measuring the resistance to JPEG-compression, adding Gaussian noise and applying linear filters. Using these processing techniques the quality of the image must be affected significantly before the watermark is removed. In this paper a method is proposed to estimate a pseudo-random spread spectrum watermark only from the watermarked image. If this estimated watermark is subtracted from the watermarked image, the watermark is removed without distorting the image significantly.