JPEG Compression Detection Based on Edge-Corner Features Using SVM

This paper focuses on the detection of JPEG compression (JC) image forensics and extracts a feature vector that composed of the Hough line, peaks, and the Harris-Stephens corner features to classify the JC and the other type images. The longest Hough line is computed by the Hough transform with the Canny line, then the coordinates of the line's endpoints would be the feature set. Also, the coordinates of the deep Hough peaks is defined as the feature set. Lastly, the coordinates of the Harris-Stephens corners would be the feature set, respectively. They are to be combined the feature vector for the JC detection. The defined feature vector is trained inSVM (Support Vector Machine) classifier for the JC detection of the forged images. The performance of the proposed JC detection is measured with the chose four types of the forged images in the experiment: unaltered, median filtering (3 × 3), averaging filter (3 × 3) and downscaling (0.9), respectively. Subsequently, the experimental items; the AUC (Area Under Curve) by the sensitivity and 1-specificity, P
_TP
at P
_FP
= 0.01, Pe (a minimal average decision error), and the classification are evaluating the performance of the proposed JC detector scheme. Thus, it confirmed that the grade evaluation of the proposed algorithm is 'Excellent (A)'.