Abstract: This paper presented a novel Content-Based Copy Detection (CBCD) scheme using Singular Value Decomposition (SVD) in the wavelet domain and early fusion for passive image forensics and Digital Rights Management (DRM). To improve the efficiency of image descriptors, multiscale singular value vectors combining global and local features of an image were exploited to generate the signature set for comparison. Local features were extracted by image partitioning and Largest Singular Value (LSV). Experimental results demonstrate the proposed algorithm not only achieves good robustness and discriminability in identifying various modified versions of an original image including geometric transformation, signal processing, image manipulation, and the combination of those but also offers improved detection performance in dealing with various rotations, shiftings, and cutting the area of an image. The proposed approach is applied to detect pirated copies of digital images in a database or Internet.

Key words: Content-Based Copy Detection (CBCD), digital image forensics, Singular Value Decomposition (SVD), Digital Rights Management (DRM), multiscale wavelet analysis

摘要： 提出一种基于小波域奇异值分解（SVD）和早期融合技术的数字图像拷贝检测算法。这种基于内容的拷贝检测模式主要面向数字图像被动式取证和数字版权管理等领域。为了提高图像描述特征的效率，算法利用多尺度小波分析提取并融合具有图像全局和局部特征的多尺度奇异值特征向量。实验结果表明,该算法不仅在识别几何变换、信号处理、图像操作处理及组合变换等不同攻击下的图像修改版本时具有较强的鲁棒性和内容辨识性，而且具有较高的检测率。算法可以用于数据库或网络环境下的数字图像盗版检测。

关键词: 基于内容的拷贝检测, 数字图像取证, 奇异值分解, 数字版权管理, 多尺度小波分析