Efficient approaches for digital image forgery detection

Abstract

Digital Image forensics is a research field which aims to validate the originality of digital images by recovering information about their history. Digital images are easily forged by various methods, which lead to change their meaning and its authenticity. This thesis aims to provide the frameworks and algorithms for the copy move forgery detection in digital images and their classification into the category of authentic and forgery images. newlineThe fundamental concept behind the Digital Image Forensics, its detection approaches, type of forgery, and its key areas along with the major issues are discussed in the thesis. It further provides the motivation to propose (efficient) copy move forgery detection methods with the use of optimization techniques and classifier. newlineA comparative study and analysis of the existing copy move forgery detection and classification based on optimization techniques has been discussed. The proposed work offers the Copy-Move based image Forgery Detection and explains how the classification is useful for forgery detection. This work discussed the Biorthogonal Wavelet Transform with Singular Value Decomposition (BWT-SVD) based feature extraction, and then it is applied to find the image forgery. The proposed Improved Relevance Vector Machine classifies the image as authentic or forgery image. The performance of this method is tested by CoMoFoD database and measured in terms of performance metrics to classify the result.A Hybrid Extreme Learning Machine with Fuzzy Sigmoid Kernel based algorithm has been designed considering the efficient image forgery detection for both copy move forgery and spliced image forgery. It is very efficient to detect forgery due to the efficient preprocessing with better transform feature extraction and optimal boundary detection. To detect the forgery, the Active Contour based Snake (ACS) method is proposed for efficient boundary detection. Then, the Contourlet Transform (CT) is introduced to extract the feature vector from the object boundary detection region.