National Institute of Technology, Srinagar, India.
Detecting Forgery in Images: A Statistical Perspective.Images have become inseparable part of our life. Almost in all fields ranging from media to evidence in courtrooms, images play an important role. Whereas digitization of images have opened new vistas in image, processing and analysis techniques that enable to extract concealed information from images that may even be beyond visual perception. However, because of the easy availability of photo editing software’s digitized images have, at the same time, become vulnerable to image tampering. Attacker may tamper the images to mislead the public, distort truth, and destroy someone’s reputation without leaving any trace. This puts authenticity of any image in doubt. Although techniques like watermarking and stenography can be used to check authenticity of an image but these techniques are no longer viable for every generated image in view of cost in terms of time and complexity. This limitation is overcome by digital image forensics. We need a reliable forensic technique that is able to act as an evidence to image authenticity. A number of forensic image authenticity techniques have been proposed. These work with varying degree of reliability. In our approach, we base our solution on the hypothesis that tampering may change underlying statistics of an image; though traces left by tampering may not be perceptible. It may be pointed out that a number of image forgery techniques exist. However, to test the proposed technique we have used two most commonly used forgery techniques Copy-Move and Splicing on images taken from two standard databases CASIA and CoMoFoD. To test the proposed hypothesis, efficacy of Grey Level Run Length Method (GLRLM) based on second order statistics has been used to detect forgery in images. The features obtained based on GLRLM have demonstrated the potential of proposed method in detecting image forgeries.