Splicing Localization in Tampered Blurred Images

Abstract

Digital images are the fastest means of transferring information in the present era, but with the ubiquitousness of advanced photo-editing tools, image forgery has become easier and more frequent. A common class of image forgery techniques is called “splicing”, in which, the forger crops a region of the first image and places it in the second image. Doing so, the difference in the blur type of different regions can leave a trace for tracking the forgery, that is, the blur type of different regions are going to be inconsistent if the source and spliced images are different in their class of blurriness. An approach to detect splicing in images is to evaluate the inconsistency of statistical features which results when splicing occurs. Considering the importance of splicing detection, various methods have been proposed. Nevertheless, the lack of public benchmark dataset for fairly evaluating the splicing detection methods is a major problem. Consequently, we were motivated to prepare a dataset for exploiting the inconsistency of blur types with the purpose of localizing splicing in forged images. We called the first version of our dataset SBU-TIDED1 (SBU Tampered Image Detection Evaluation Dataset). Furthermore, we have explored the features used in blur type detection. A new set of features is proposed which leads to accuracy enhancement. It is apparent form the experimental results that the proposed features are efficient for detecting and classifying two blur types, namely, out-of-focus and motion blur.