Copy move image forgery detection through modified sift and cnn Feature extraction

Abstract

The significance of ensuring the authenticity of images is newlineincreasing with the ready availability of image-editing software. Image newlineprocessing techniques for detecting digital picture forgery are gaining newlinesignificance and popularity over recent days. Detection of criminal cases may newlinebe hindered by hiding evidence with picture fabrication with the sophisticated newlinesoftware that is currently available with the advancements in image newlineprocessing technology. Digital image forging is the technique of generating newlineforged images by modifying the original photographic images. In copy-move newlineforgery, particular elements of an image are duplicated, or an item in the newlineimage is hidden by copying a section of an image and pasting it on another newlinepart of the same image. During tampering, the tampered regions frequently newlineundergo post-processing or geometric techniques to make the forgeries look newlineinconspicuous and real. The challenges in forgery detection are increased by newlinethe realistic doctored images produced by post-processing techniques. The newlinecrucial evidence in copy-move forgery is the similarity between the source newlineand the tampered regions. The best mechanism for identifying copy-move forgery is to newlinesegregate the image into overlapping square blocks. This segregated image newlinecan further be represented as blocks with the help of Discrete cosine newlinetransform (DCT). Traditionally, Principal component analysis with Gaussian newlineRadial basis function (RBF) was used to improve the efficiency of feature newlinematching. This research introduces an enhanced Scale-invariant feature newlineTransform (SIFT) known as RootSIFT. This technique is used to identify and newlineaccurately point to the tampered area of the image. Experimental observations newlineindicate positive figures in parameters such as forgery location accuracy, newlinedetection reliability and time complexity. Moreover, the F1 score reaches a newlinemaximum of 99.52 with the proposed technique, which is the highest newlinecompared to other traditional methodologies while testing the MICC-F220 newlinedataset. The second part of this work introduces an ADaptive Scale- newlineInvariant Feature Transform (ADSIFT) algorithm for detecting copy-move newlineforgery. Previous Keypoint-based detection methods were robust against newlinelarge-scale geometric transformation but lacked in producing the necessary newlinenumber of keypoints. To overcome this defect, ADSIFT proposes newlineincorporating a rescaling factor and a gamma factor for feature matching and newlinecontrast threshold. This will produce adequate keypoints despite low contrast newlineand smooth or small regions of noiseless or noisy images. newline newline