Design and Analysis of Efficient Methodologies for Forgery Detection in Forensic Images

Abstract

Image forgery has become an important field in the recent past due to increasing number of newlinecomplaints about tampering of images. The image tampering is carried to hide some information newlinefrom the image or to change the meaning of the image. In this work, significance of tempering newlinedetection and its effects on the decision making are discussed. A variety of methods used to newlinetamper an image have been outlined. Methods used in forgery detection like active methods and newlinepassive methods are presented. One of the popular methods, namely, Copy paste method which newlineis predominantly used by imposters is presented. Copy-paste methods based on block based newlinemethods and key point descriptors are outlined. Other popular techniques used in the forgery newlinedetection like local binary pattern (LBP), discrete cosine methods (DCT) and singular value newlinedecomposing (SVD) are also explained. newlineThree phases of research work, namely, SIFT and RANSAC algorithm to detect forgery due to newlinecopy paste from the same image, forward quantization method to detect if image compressed and newlinedecompressed after forgery and forward quantization classification to detect if an image has any newlineforgery, has been presented in the thesis. newlineIn the first phase, the improvised Scale-Invariant Feature Transform (SIFT) and Random sample newlineconsensus (RANSAC) algorithm is introduced. The improvised algorithm is developed in this newlinework to extract the tampered region from forged images. The improvised algorithm is presented newlinewith several details. A data set of one hundred images with good and tampered images has been newlineused in the simulations to measure the performance of improvised algorithm. Several metrics like newlinepercentage area of tampered region with respect to overall size of the image, absolute difference newlineor signal to noise ratio are derived for each of the forged images in the dataset used. newlineii newlineIn the second phase, another new algorithm has been developed based on a novel approach of newlinegoodness of fitness for forward quantization noise. The image needs to be first categor