Blind Detection of Digital Photomontage using Higher Order Statistics

The advent of the modern digital technology has not only brought about the prevalent use of digital images in our daily activities but also the ease of creating image forgery such as digital photomontages using publicly accessible and userfriendly image processing tools such as Adobe Photoshop. Among all operations involved in image photomontage, image splicing can be considered the most fundamental and essential operation. In this report, our goal is to detect spliced images by a passive-blind approach, which can do without any prior information, as well as without the need of embedding watermark or extracting image features at the moment of image acquisition. Bicoherence, a third-order moment spectra and an effective technique for detecting quadratic phase coupling (QPC), has been previously proposed for passive-blind detection of human speech splicing, based on the assumption that human speech signal is low in QPC. However, images originally have non-trivial level of bispectrum energy, which implies an originally significant level of QPC. Hence, we argue that straightforward applications of bicoherence features for detecting image splicing are not effective. Furthermore, the theoretical connection between bicoherence and image splicing is not clear. For this work, we created a data set, which contains 933 authentic and 912 spliced image blocks. Besides that, we proposed two general methods, i.e., characterizing the image features that bicoherence is sensitive to and estimating the splicinginvariant component, for improving the performance of the bicoherence technique. We also proposed a model of image splicing to explain the effectiveness of bicoherence for image-splicing detection. Finally, we evaluate the performance of the features derived from the proposed improvement methods by Support Vector Machine (SVM) classification on the data set. The results show a significant improvement in image splicing detection accuracy, from 62% to 72%.