Friday, September 22, 2017

New M.Tech. Thesis Submitted from computer science

by Geetika Gupta 

Today’s technological world is characterized by the presence of digital images that are vastly spreading all over the world and various image editing tools are available that have made very easy for a common user to create a digital forgery. Images are used as legal evidence in many fields and if these images are corrupted, then it will generate a crucial issue. A copy-move forgery is performed in images either for cloning, or concealing some object. In both situations, an image lost its integrity and hence detecting this type of forgery has becomes very challenging problem as the replicated zone arrives from the same source image. The primary objective is to develop a mechanism that will verify the credibility of an image and discover the duplicated areas in it. This research work proposes a passive procedure in which no information about the original image is involved. Firstly, the grayscale image is split into overlapping blocks and then hybrid (principal component analysis and histogram of oriented gradients) techniques are used for extracting the features. At last, the feature vectors are sorted lexicographically and replicated areas are matched. To compare the performance, an exact match method and proposed method was implemented. The proposed method reduces the false matches or discontinuities that have occurred in exact match method by using a preset offset threshold value. The qualitative inspection shows that the proposed method gives the best results and is more resilient under two different attacks (salt and pepper noise and filtering attack) as compared to exact match method. Moreover, the quantitative analysis shows that the proposed method gives the better results in terms of three different metrics, namely, precision, recall and accuracy. It achieves 86.96% precision, 80% recall and 84% accuracy as compared to exact match method.