A novel blind watermarking scheme based on neural networks for image

摘要

In this paper, a novel blind watermarking scheme based on the back-propagation neural networks (BPNN) for image is presented. First, the convolutional codes encoding is used to refine the watermark for increasing robustness of the scheme. BPNN is developed to memorize the relationships between the wavelet selected samples and a processed chaotic sequence. With wavelet domain of original image being divided into watermarking blocks, then several different BPNN models of selected watermarking blocks are trained simultaneously to form certain relationships, which are employed for embedding the coded watermark bit stream. Compared with conventional watermarking, the proposed scheme based on the trained BPNN models modifies only a small amount of image data such that the distortion on original image is imperceptible. Experimental results demonstrate the high robustness of the proposed scheme against common signal processing.