Median filtering detection using variation of neighboring line pairs for image forensic

Robust feature vector is required for a design of the median filtering (MF) detection in the digital forgery image. This paper presents a new feature vector that is composed of two kind variations between the neighboring line pairs (the row and column directions). In the proposed scheme for the variation, the one which is defined from a gradient difference of the intensity values between the neighboring line pairs, and the other one is defined from a coefficient difference of Fourier Transform (FT) between the neighboring line pairs too. Subsequently, the constructed 10-dimensional feature vector is trained in a SVM (Support Vector Machine) classifier for MF detection in the altered images. The proposed variation-based MF detection scheme that has the same feature vector length like as the MFR (Median Filter Residual) scheme, thus compared to the performances of MFR. As a result, in the measured performances of the experimental items, the AUCs (Area Under ROC (Receiver operating characteristic) Curve) by the sensitivity (Ptp: the false positive rate) and 1-specificity (Pfp: the true positive rate) are above 0.9, and the classification ratios are above 0.9 too. Pe (a minimal average decision error) is ranged from 0.003 to 0.027, and Ptp at Pfp=0.01 is ranged from 0.965 to 0.996. In spite of the 10-D shorter length of the feature vector, it is confirmed that the grade evaluation of the proposed variation-based MF detection scheme rated as `Excellent (A)´.