Detection of object-based manipulation by the statistical features of object contour

Object-based manipulations, such as adding or removing objects for digital video, are usually malicious forgery operations. Compared with the conventional double MPEG compression or frame-based tampering, it makes more sense to detect these object-based manipulations because they might directly affect our understanding towards the video content. In this paper, a passive video forensics scheme is proposed for object-based forgery operations. After extracting the adjustable width areas around object boundary, several statistical features such as the moment features of detailed wavelet coefficients and the average gradient of each colour channel are obtained and input into support vector machine (SVM) as feature vectors for the classification of natural objects and forged ones. Experimental results on several videos sequence with static background show that the proposed approach can achieve an accuracy of correct detection from 70% to 95%.