基于多特征融合的图像拼接检测

针对数字图像篡改的常用手法图像拼接,提出了一种基于多特征融合的被动盲取证算法来检测图像拼接.算法通过分析图像相位一致性和纹理特征,采用二维经验模式分解将图像分解到固有模态函数域,得到三类特征值.利用这三类特征值,采用支持向量机作为分类器,建立一个预测模型,对图像是否经过篡改进行判定.选用标准图像拼接库对该算法进行测试.实验结果表明:与采用双相干谱作为分类特征的算法相比,该算法具有更高的识别率.

Image splicing detection using multi-features amalgamation

Image splicing is a usual way to tamper digital image.This paper proposes a blind and passive forensic scheme based on multi-features amalgamation to detect image splicing.Three kinds of features of an image are got by analyzing its phase congruency and texture features,and it is disassembled into the Intrinsic Mode Function domain through Bidimensional Empirical Mode Decomposition.Utilizing the features,a forecast model is constructed with a support vector machine as the classifier to judge whether the image is forged.The proposed scheme is evaluated with the standard spliced image dataset.The experimental results indicate that this scheme has higher detection accuracy than the algorithm using the bicoherence magnitude.