Brightness distribution based image tampering detection

Reports confirming malicious manipulation of digital images are increasing nowadays. Law enforcement agencies are looking for solutions capable of authenticating digital images. In this paper, we have proposed a method that detects tampering involving facial regions by comparing the dissimilarity between the brightness distributions. We have considered two cases of tampered images; images with three faces and images with two faces. For images with three faces, distance comparison between histograms of three faces is carried out for detecting tampered face. For images with two faces, the distance between two faces is checked against a threshold to detect if the image is tampered. Both bin-by-bin measures and cross-bin measures are computed as distance measures between brightness distributions. Experiments conducted on images with three faces reveal that the three bin-by-bin measures such as Minkowski-form distance, Chi-square distance and Bhattacharyya distance give a true positive rate of 56.76%, 59.46% and 48.65% respectively, whereas cross-bin measures such as Earth Movers Distance (EMD), Match distance and Kolmogorov-Smirnov distance give lower true positive rates of 32.43%, 32.43% and 37.84% respectively. Results of threshold based approach for images with two faces show that the three bin-by-bin measures such as Minkowski-form distance, Chi-square distance and Bhattacharyya distance give a true positive rate of 63.16%, 63.16% and 60.53% respectively, whereas cross-bin measures such as EMD, Match distance and Kolmogorov-Smirnov distance give lower true positive rates of 50%, 50% and 55.26% respectively.