A SUPER-SIFT APPROACH FOR COPY-MOVE FORGERY DETECTION

Abstract

Today’s technological era is described by the outspread of digital images. They are the most ordinary formation of conveying information whether through newspapers, internet, books, magazine, scientific journals or social media. They are used as a powerful proof against various crimes, frauds and as an evidence in various situations. With the evolution of image processing in past few years and many other image editing software, capturing, creating or altering images according to our perspective has become very simple and available. There are several kinds of image tampering like copy-move forgery, image enhancement, image splicing, image morphing, image retouching whereas copy-move forgery is the most frequent and trendy manipulation of digital images. In copy move forgery here, a part of particular image is copied and then pasted into that same image with the motive of veiling some important object or displaying a fictitious scenario. Because the duplicate or in other terms the copied portion comes from the same image, All the image properties like texture, noise, resolution, brightness, contrast will be suited with the original portion of the image making it more difficult for the experts to distinguish and detect the alteration. There are mostly two kinds of forgery detection techniques one is block based method and the other is based on key points. In past few years feature based approach like SIFT gain attention of researchers in the field of image forgery detection. I proposed a SUPER-SIFT method for copy move forgery detection. This work improves the fundamental concept of SIFT algorithm which is Feature Extraction. We use SISR for improving the quality of image. The proposed work consist of three main tasks, firstly we preprocess the input image with SISR algorithm to get a high resolution image. Then on high resolution image we apply SIFT algorithm for keypoint detection. After that we apply a fast potential based hierarchical agglomerative clustering method on the output of previous step to filter out the false matches and to groups the key points that have the same affine transform. On the basis of number of key points in a particular cluster, it can be said that the image having forgery or not. The experimental outcome shows that the proposed approach for the detection of copy-move forgery is efficient and powerful even when the copied portion undergoes various transformations like rotation, shearing, scaling or other post processing like adding noise and blurring.