Image forgery detection based on energy probability

The development and widespread deployment of easy to use image editing tools has increased imageforgery operations in recent years. Many methods have been proposed in the literature to detect the copy move operation, one of the most widely used forgery techniques. Methods exhibit differences from one another by the way they used to represent sub-blocks in the image. One of the most important problems for all techniques in use is to decide the threshold value to use in order to detect similarity successfully.

In this work, DCT energy probability values of overlapping sub-blocks are used to construct feature vectors of size 1×10. Using a similarity ratio instead of trying to decide a static threshold is an advantage of the method compared to similar techniques. The proposed method yields better results compared to other DCT based methods under attacks (such as JPEG compression and blurring) to hide clues of forgery.