基于SVM的多通道鲁棒图像零水印算法

A Robust Image Zero-Watermarking Algorithm Based on Multi-Channel and SVM

摘要: 本文基于多通道水印模型和SVM，提出了一种鲁棒零水印算法。该算法将DWT域划分为多个水印通道，基于水印和混沌密钥确定的DWT系数构造样本集，并对每一通道进行SVM训练，从而生成决策函数。水印检测阶段，通过各通道的决策函数提取各通道的水印，最后通过各通道水印和估算方法计算最终的鲁棒水印。由于该算法方案并没有真实嵌入水印信息，图像没有任何失真。另外，多通道方案使得该算法具有强鲁棒性。

Abstract: This paper proposes a robust image zero-watermarking scheme based on a multi-channel watermarking model and Support Vector Machine (SVM). In the scheme, the DWT domain of the original image is divided into many non-overlapping watermarking channels, and SVM training sets of all watermarking channels are built based on robust watermark and DWT coefficients decided by a logistic chaos key. By SVM training, we obtain decision functions of all watermarking channels. During the watermark detecting stage, watermarks of all watermarking channels are extracted by the decision functions, and the final robust watermark is worked out by an estimation function. With the proposed algorithm, the watermarked images are perfect without any distortion since we don’t embed any information into the original image, and they can resist various intensive watermarking attacks.