基于SVR和DCT的数字图像水印算法

针对彩色图像,提出了一种利用支持向量回归机（support vector regression, SVR）将水印信息嵌入到图像亮度分量的离散余弦变换（discrete cosine transform, DCT）域中的算法。主要思想是应用彩色图像的YCbCr彩色空间,将亮度分量分块后做离散余弦变换,在DCT块中选择中频系数,利用SVR很好的学习和泛化能力,建立中频系数与其邻域系数之间的关系模型,然后根据中频系数和SVR模型预测输出值,调整中频系数嵌入水印。提取水印时重新训练SVR模型,但不需要原始载体图像,实现了水印的盲检测。实验结果表明,该算法具有良好的不可感知性,对色度变化、低通滤波、模糊、锐化、亮度、对比度增强、添加噪声以及几何变换等攻击均具有较强的鲁棒性,尤其对JPEG压缩具有很强的抵抗能力。

Watermarking algorithm for digital images based on SVR and DCT

A novel algorithm embedded watermark information into the discrete cosine transform (DCT) domain of the brightness component of the color image is proposed by means of support vector regression (SVR). Using the YCbCr color space of the color image, the brightness component is divided into blocks and each block is done discrete cosine transform. The intermediate frequency coefficient is chosen in the domain and the relation model between the coefficient and its neighborhood is constructed by utilizing the good learning ability and generalization ability of SVR. According to the intermediate coefficient and output predicted by the SVR model, the watermark is embedded by adjusting the coefficient. Furthermore, the SVR model is reconstructed during the watermark extraction. The algorithm proposed is a blind detection due to the process of extracting watermark without the original color image. In addition, the experimental results further demonstrate that this algorithm has the good imperceptibility and the strong robustness for chrominance transformation, lowpass filtering, blurring, sharpening, brightness and contrast enhancement, noise addition and geometric transformation, especially for JPEG compression.