Shift recompression-based feature mining for detecting content-aware scaled forgery in JPEG images

Abstract

Content-aware image resizing, also known as image retargeting, seam carving, content-aware scaling, is originally proposed to automatically remove the paths of least importance, known as seams, to reduce image size or insert seams to extend it, in order to display images without distortion on various media especially on mobile devices, such as smartphones and PDAs. Content-aware scaling also allows removing entire objects from photographs without observed clues, and hence it has been used to tamper images. Due to the ubiquity of JPEG images on various mobile devices, it is increasingly necessary to authenticate these JPEG images for legitimate purposes. To detect the content-aware-based forgery in JPEG images, in this paper, we merge shift-recompression-based characteristic features in spatial domain and shift-recompression-based neighboring joint density in DCT domain together; an ensemble classifier is used to discriminate forged JPEG images from intact JPEG images. We also transfer other popular JPEG-based steganalysis methods to detecting the forgery. Experimental results show that steganalysis methods are effective in detecting context-aware-based JPEG forgery and our method is superior to other compared detection methods.