一种新的基于DCT域系数对直方图的图像篡改取证方法

A novel image forensic method based on coefficient-pair histogram in DCT domain

提出了一种基于离散系数变换(DCT)域系数对直方图特征的图像篡改取证方法。首先对图像进行DCT,并在给定的阈值下,对变换后的DCT系数进行系数对直方图化;而后对直方图化后的值进行归一化处理,再用主成分分析(PCA)对上述数据进行降维处理取得系数对直方图特征;最后将真实图像和篡改后图像的系数对特征用支持向量机(SVM)进行分类识别。实验结果证明,和现有的一些算法相比,提出的方法计算复杂度低,在CASIA v1.0平均识别率为97.92%,CASIA v2.0平均识别率为91.20%,对未压缩图像和压缩图像的拼接篡改都具有良好的识别性能。

Digital images can be forged easily with today＇s widely available image processing software. One of the most common practices in image tampering involves cropping a patch from a source and pas- ting it onto a target. This task,when images are used as evidence to influence or decide judgment such as in a court of law, can be crucial. A novel image forensic method based on coefficient-pair histogram in discrete cosine transform （DCT） domain is proposed. In the method, firstly, the image is transformed by DCT, then with the given threshold, coefficienvpair histogram is computed for the DCT coefficients, and normalization processing is excuted on the coefficient-pair histogram. After that, principal component a nalysis （PCA） is used to reduce the dimension of the above data to get the final image features. Lastly, support vector machine （SVM） is exploited to classify the authentic and spliced images through training the feature vectors of them. In CASIA vl. 0, the recognition rate reach 97. 92~/00, and it is 91.20%o in CASIA v2.0. The experimental results show that compared with some existing methods, the proposed approach has low computing complexity and good recognition performance in both uncompressed and compressed images.