Abstract: How to discriminate between photorealistic computer generated images and real photographic images captured with a digital camera has become an important component of digital image forensics.To solve this problem，this paper presents a novel image authentication approach using spectral correlation between color planes in an image for forensic analysis.In single-CCD imaging systems，only one color component per pixel is captured，the missing two color values at the pixel are interpolated by CFA（Color Filter Array）.Demosaicked images by CFA interpolation exhibit higher spectral correlation between different color planes.In the suggested method，statistical spectral correlations between color components are exploited using discrete wavelet transform and normalized cross correlation to verify photographic images.Experimental results tested on the de facto standard Columbia dataset demonstrate that these features are effectual in capturing the differences between two kinds of images and the proposed method achieves promising accuracy in passive image authentication.

Key words: digital image forensics, spectral correlation, passive image authentication, wavelet decomposition

摘要： 如何区分逼真的计算机生成图像和真实的自然图像，是数字图像取证领域的一个重要研究方向。提出了一种基于谱间相关性的图像真伪鉴别算法。在单CCD数码成像过程中，每个像素只采集单一颜色值，缺失的颜色值通过颜色滤波阵列插值获得，而基于颜色滤波阵列插值的去马赛克方法会引起彩色图像三个颜色分量之间较高的谱间相关性。算法利用小波变换和标准互相关系数提取颜色组件谱间相关性，作为区分和识别特征。通过在标准图像库上的实验测试，表明所提取特征有效捕获了二类图像的差别，并具有较高的检测率。

关键词: 数字图像取证, 谱间相关性, 被动式图像鉴别, 小波分解