摘要：

针对多重水印技术如何确定最优的嵌入区域, 以实现多个水印之间的权衡问题, 提出一种基于多通道嵌入的彩色图像多重水印算法。当嵌入水印时, 选择在彩色图像R、G、B 3个通道嵌入不同的水印图像, 目的是嵌入更多的水印信息。通过提升小波变换(LWT)和非下采样剪切波变换(NSST)提取宿主图像每个通道中的方向性特征, 确定水印的嵌入位置, 以提高对常规攻击和几何攻击的鲁棒性。通过将加密的水印有效信息嵌入到具有良好稳定性的奇异值矩阵中, 保证水印的提取效果。实验结果表明, 嵌入水印后的图像峰值信噪比(PSNR)在43 dB以上, 所提算法具备良好的不可见性, 对嵌入水印后的图像进行高强度的旋转、滤波、剪切等攻击, 提取水印的归一化系数(NC)值在0.95以上, 其中, 滤波攻击对应的NC值可以达到0.99以上, 水印信息提取完整、清晰可辨, 表现出较强的鲁棒性。同时, 该算法可以1次嵌入3个水印图像, 具有更高的嵌入容量。因此, 与现有的多重水印算法相比, 该算法的不可见性、鲁棒性和嵌入容量均有较大的提升。

关键词: 多重水印, 提升小波变换, 非下采样剪切波变换, 鲁棒性, 嵌入容量

Abstract:

To address the problem of determining the optimal embedding area in multiple watermarking technologies and to achieve a trade-off between multiple watermarks, a multi-channel embedding-based color image multiple watermarking algorithm is proposed. Different watermark images are embedded in the R, G, and B channels of a color image to embed additional watermark information. The directional features in each host image channel are extracted using the Lifting Wavelet Transform (LWT) and Non-Subsampled Shearlet Transform (NSST) to determine the watermark embedding position for improving robustness against conventional and geometric attacks. Watermark extraction is guaranteed by embedding the encrypted effective watermark information into a singular value matrix with good stability. The experimental results demonstrate that the Peak Signal-to-Noise Ratio (PSNR) of the image after embedding the watermark is above 43 dB. Furthermore, the algorithm has good invisibility, high-intensity rotation, filtering, and shearing. Additionally, other attacks are performed on the image after the watermark is embedded for watermark extraction. The Normalization Coefficient (NC) value is above 0.95, and the NC value corresponding to the filter attack exceeds 0.99. Watermark information extraction is complete and clear, showing strong robustness, and three watermark images can be embedded with a higher embedding capacity. In summary, the invisibility, robustness, and embedding capacity of the proposed algorithm are significantly improved in contrast to those of multiple existing watermarking algorithms.

Key words: multiple watermarking, Lifting Wavelet Transform(LWT), Non-Subsampled Shearlet Transform(NSST), robustness, embedding capacity