Image Forgery Forensics Based on Manual Blurred Edge Detection

In this paper, a novel image forensics method is proposed to detect manual blurred edges from a tampered image. Firstly, the image edges are analyzed by using non-subsampled contourlet transform. Then the differences between the normal edge and the blurred edge are extracted by researching phase congruency and prediction-error image. After that, the features are used to train the SVM, by which the blurred edges can be distinguished. Finally, the local definition is introduced to indicate the differences between the manual blur and defocus ones. Experimental results show that this method can detect possible blurring in images and locate the tampering boundary with a relative high accurate rate.