摘要
抗几何攻击的鲁棒图像水印设计是目前水印技术研究的难点和热点之一.文中分析了图像的Bandelet变换特性,提出了一种以图像特征点矢量集为特征向量的回归支持向量机(Support vector regression,SVR)和第二代Bandelet变换的抗几何攻击图像水印算法,采取的主要方法包括:1)在Bandelet变换提取的刻画图像局部特征的几何流系数上,采用奇偶量化嵌入水印;2)利用Harris-Laplace算子从归一化的含水印图像中提取具有几何形变鲁棒性的图像特征点,构造特征点矢量集作为特征向量,应用回归支持向量机对几何变换参数进行训练学习;3)水印检测时,先利用SVR训练模型得到待检测图像所受几何攻击的参数并作几何校正,然后通过奇偶检测器盲提取水印.仿真实验表明,所提出的水印算法不仅具有良好的透明性,而且对常规图像处理、一般性几何攻击和组合攻击均具有良好的鲁棒性.
Image watermarking against geometric attacks is the hotspot and challenging point in the present research on watermarking. This paper analyses the characteristics of the second generation Bandelet translation, and then proposes a novel Bandelet-domain watermarking scheme against geometric attacks based on support vector regression （SVR） with the vector of feature points. The proposed scheme includes three important techniques： 1） the Bandelet based image directional flow coefficient which depicts the characteristic of the image is used to embed watermarking bits with odd-even quantization; 2） the feature points robust to geometric deformation are extracted from the watermarking image using Harris-Laplace operator, and used as the eigenvectors to train the SVR model; 3） during detection, the parameters of geometric attacks are obtained using the well trained SVR, which is used for resynchronization, then an odd-even detector is used to extract the watermark blindly. Experiment results show that the proposed scheme is well transparent and is not only robust to common image processing but also robust against some geometric attacks as well as some combined attacks.
出处
《自动化学报》
EI
CSCD
北大核心
2012年第10期1646-1653,共8页
Acta Automatica Sinica
基金
广东省自然科学基金(S2012010010004)
网络与数据安全四川省重点实验室2011开放项目
广东省高等院校高层次人才项目资助~~
关键词
图像水印
几何攻击
BANDELET变换
回归支持向量机
奇偶量化
Image watermarking, geometrical attacks, Bandelet transform, support vector regression （SVR）, odd-evenquantization
