基于核判别分析与证据理论的图像伪作分层融合检测

Image Forgery Detection Based on Kernel Discriminant Analysis and Evidence Theory

现有图像伪作融合检测算法一般直接采用特征融合或决策融合技术,普遍存在算法不易扩展或检测准确率不理想等问题。在综合利用原始图像固有特征和篡改所引入特征的基础上,探讨了一种基于特征融合和决策融合的分层融合框架,并实现基于核判别分析(kernel discriminant analysis,KDA)和证据理论的图像伪作检测算法。该算法包含粗分类和细分类两阶段。在粗分类中,利用原始图像固有特征,采用KDA技术实现特征融合,输出结果为原始图像、篡改图像和待定图像三种类别。在细分类中,利用篡改操作所引入的特征,采用证据理论进行决策融合,实现对待定图像的进一步分类。实验结果表明,该算法能有效地检测模糊操作、重采样操作、JPEG压缩以及多种篡改组合操作。

Information fusion has become a new hotspot in image forensics. In most of the currently presentedmethods, the related image forgery detection is usually carried out by using feature fusion or decision fusion. A hier-archical fusion consisted of feature fusion and decision fusion is proposed to improve performance accuracy in imageforgery detection. Firstly, multiple inherent features of a suspected image are extracted and fused by using kerneldiscriminant analysis （KDA） , then classified into forgery, non-forgery or undetermined class. Finally, for the un-determined image, tampering features are extracted and fused by using evidence theory to fulfill detection task. Ex-perimental results show the feasibility of the proposed method for image forgery detection.