Revealing digital fakery using multiresolution decomposition and higher order statistics

With the advance of digitization and digital processing techniques, digital images are now easy to create and manipulate, and leave no clues of artificial evidence. There are some known digital fakery for images, e.g., computer graphics (CGs) and digital forgeries. As valid records of natural world, natural images, i.e., photographic images, are no longer believable. In this paper, a detection scheme for natural images and fake images is proposed. Features are first extracted using multiresolution decomposition and higher order local autocorrelations (HLACs). The support vector machines (SVMs) are then used to differentiate the natural and fake images. Because the inner product between features can be obtained directly without computing features, it can be integrated into SVM, and the computation complexity is decreased. Experiments show that the proposed detection scheme is effective, demonstrating that the proposed statistical features can model the differences between natural images and fake images.