Blind image forensics using reciprocal singular value curve based local statistical features

Abstract

In this article, passive contrast enhancement detection technique is presented using block based reciprocal singular value curve features. Contrast enhancement operation changes the natural statistics of the image and variation in singular value curve is exploited for constructing the feature vector for forgery detection. Various statistical features using reciprocal singular value curve are extracted after multilevel 2-Dimensional wavelet decomposition. Fisher criterion is employed to choose the most discriminating and to discard the redundant features. Experimental results are presented using gray scale, G component and C b image database and support vector machine classifier. Robustness against anti-forensic algorithm and JPEG compression is also presented. The algorithm outperforms all the existing feature based blind contrast enhancement detection methods in terms of detection accuracy.