基于神经网络的复合变换域视频水印算法研究

Research on Video Watermark Algorithm in Composite Transform Domain Based on Neural Network

针对目前视频水印算法存在的鲁棒性较差,可靠性较低等问题,提出了一种结合神经网络将二值水印嵌入到经过离散小波变换(DWT)和离散余弦变换(DCT)后的宿主视频中的新方法;为使算法具有更好的不可见性、鲁棒性和实用性,利用三层RBF神经网络训练出水印嵌入强度,在视频中自适应嵌入水印;该方法是对宿主视频进行DWT处理,再对逼近子图LL进行DCT处理,通过修改DCT系数嵌入水印信息;在嵌入之前对二值水印进行了Arnold变换来加密;通过实验结果中PSNR与NC的值表明,算法具有很强的抗攻击和承受帧删除、帧平均等操作的能力,不可感知性好,鲁棒性明显优于一般的嵌入算法。

In view of the problems that the current video algorithms have poor robustness and lower reliability, a new scheme of embedding watermarking into video based on neural network, discrete wavelet transform （DWT） and discrete cosine transform （DCT） was proposed in this paper. In order to improve the invisibility, robust and its usefulness of the algorithm, use the three--layer neural network training to get the embedded strength and embed watermark into video adaptively. The method is to do DWT for the video, then do DCT for LL. Embed the watermarking information in the video through modifying the DCT coefficients. Before embedding, enerypt the binary watermarking by making Arnold transform. The value of PSNR and NC of the experimental results show that the new algorithm has strong ability for the attack and to bear frame dropping and frame averaging. It also has the good invisibility and robustness, and the algorithm is better than the usual embedded method.