多尺度特征耦合双分类器的图像伪造检测算法

Image authenticity detection algorithm of multi-scale feature and double classifier

为解决当前图像伪造检测技术仅局限于单一伪造形式的检测,难以适应各种复杂的组合篡改识别,使其识别精度与通用性能不佳等不足,提出多尺度特征提取耦合双分类器的图像伪造检测算法。分别利用Curve-let变换、Gabor变换、LBP(local binary pattern)与DCT(discrete cosine transform)变换采集输入图像的特征信息,融合这些提取特征,形成图像的多尺度特征;引入隐马尔科夫和支持向量机,设计双分类器的真伪决策模型,将多尺度特征视为识别依据,利用双分类器决策出真实图像和篡改图像。实验结果表明,与当前伪造检测前算法相比,所提算法具有更高的检测精度与鲁棒性,能够有效地对复制区发生旋转、模糊和噪声的复制-粘贴和拼接伪造完成精确检测。

Current image forgery detection technologies only complete the detection of a single form of forgery,and they are difficult to adapt to a variety of complex combinations of tampering identification,the recognition accuracy and the general performance are poor.To solve the problems,a method of multi-scale feature extraction and double classifier was proposed.Four kinds of feature extraction methods were used to extract the feature information of the input image,respectively.Multi-scale features of image information were formed according to the advantages of each feature extraction method.HMM and SVM were introduced,and the double classifier decision model was designed to extract the multi-scale features as the basis for judging.The double classifier was used to determine the real image and the tampered image.Experimental research shows that compared with the previous algorithms,the proposed algorithm can effectively rotate JPEG compression,blurring and noise on the replication region of the copy paste forgery and splicing detection,and it has higher detection accuracy and better robustness performances.