Accurate and Robust Localization of Duplicated Region in Copy-Move Image Forgery

Abstract

Many studies in image forgery detection have demonstrated the importance of reviewing different types of the forgeries and possible detection methods. Region duplication is an efficient operation to create image manipulation. In this type of tampering a region in an image is copied and moved onto a different area of the same image. Post processing approaches, such as possible geometrical and illumination adjustments can be applied in the duplication region to create a consistent image and to hide the forgery. The local feature extraction methods have powerful recognition and point matching applications. Since, there is a duplicated area in the copy move forged images; the local feature extraction methods can present a good outcome in detecting the forgeries. There are traditional methods in this area that used feature extraction advantages to detect the copy-move forgery in images. However, the methods mostly evaluate their accuracy in a small size dataset and without comparing with other existing methods. Moreover, locating the duplication area is the second necessarily step, after detecting and distinguishing the forged images and non-forged ones. This step is not investigated enough in the literatures. In this thesis, the use of feature extraction methods in both detection and localization stages is studied. The efficiency of presented method was verified in a large dataset including different combinations of affine transform operation. The final results were compared with the existing methods, thereby revealing more accuracy in the detection of such image forgery.