Passive Forensics Based Digital Image Forgery Detection Techniques

Abstract

Digital images are an essential part of our daily lives. They are beneficial in fields such as education, social media, newspapers, magazines, and others. Nowadays, images are the primary mode of communication. Images can be easily forged due to the rapid advancement of digital image technology and the availability of a wide range of digital image forgery devices. As a result, the main issue is the authenticity of digital images. The requirement to authenticate digital images created opportunities for an exciting field of research that focused on the development of algorithms for forgery detection. Thus, digital image authentication is critical, and the current field of study seeks to validate digital image authenticity. The proposed research aims to create a passive algorithm that detects two types of image forgery: Copy-move Forgery (CMF) and Image Splicing Forgery (ISF). In the present research, a passive strategy for forgery detection is used, which does not require any previous image data and is thus also known as the “blind strategy”.This research proposes an improved method for detecting CMF that is both efficient and sophisticated. Even though CMF is a prevalent type of forgery, detecting it is difficult because the copied part belongs to the same image and can thus exhibit similar features to that of the image. Thus, in the proposed work, block-based technique i.e. Adaptive Over-segmentation (AS), and keypoint-based techniques i.e. Accelerated KAZE (AKAZE), and Scale-Invariant Feature Transform (SIFT) are combined to detect CMF, which makes the proposed algorithm more computationally efficient and accurate for the detection and localization of single as well as multiple forgeries. The experimental results show that the proposed technique is robust against various attacks i.e. rotation, scaling, JPEG compression, and noise addition, and proved better when compared with the other existing techniques with an improved F2 score of 99.80%, 99.81%, 99.35%, and 99.82% on benchmark datasets i.e. Image Manipulation Dataset (IMD), MICC-F220, COVERAGE, and GRIP, respectively.Further research in this thesis is being conducted to detect ISF, which is also one of the most commonly used image manipulation techniques. In the proposed method, Markov features from both the Local Binary Pattern (LBP) and Discrete Wavelet Transform (DWT) domains are extracted and combined to efficiently detect ISF. For detecting ISF, it is crucial to capture the forgery-introduced artifacts, as image splicing produces sharp edges in a forged image. Moreover, the edges introduced by forgery differ from their neighbors, and thus the relationships between the spliced and original parts can be used to expose image forgery. To describe these relationships, the proposed method makes use of the Markov Transition Probability Matrix (TPM). The experimental results performed on six datasets indicate that the proposed approach offers better results than the existing techniques in terms of various performance metrics.A hybrid approach based on Discrete Fractional Cosine Transform (DFrCT) and LBP is proposed in this thesis to detect copy-move and splicing forgeries simultaneously. The additional parameter i.e. fractional parameter of DFrCT is utilized to improve the accuracy and LBP is used to highlight the tampering artifacts effectually. Additionally, localization is performed on both the copy-move and spliced images to localize the image's duplicated areas. The efficacy of the proposed scheme is confirmed by extensive simulations on six benchmark datasets, namely CASIA v1.0, GRIP, CASIA v2.0, IMD, COVERAGE, and Columbia, which achieved accuracy rates of 99.67%, 99.23%, 99.76%, 98.81%, 95%, and 98.17%, respectively, that surpasses existing techniques.A Contrast Limited Adaptive Histogram Equalization (CLAHE) based Convolutional Neural Network (CNN) model is presented for effectively solving the issue of CMF detection. The CLAHE algorithm makes the hidden features of the image visible, as some of them are hard to detect in CMF. The proposed work primarily focuses on improving the performance parameters of the forgery detection, which are superior to the existing techniques with reduced False Negative Rate (FNR) of 0.0132, 0.0179, 0.0000, and 0.0010 on MICC-F220, GRIP, IMD and MICC-F2000 datasets, respectively. Also, the robustness of the proposed technique is demonstrated against several attacks like scaling, noise addition, JPEG compression, and rotation. Furthermore, statistical analysis tests such as Analysis of Variance (ANOVA) and cross-dataset performance are used to validate the effectiveness of all the proposed strategies. Future work could be devoted to expanding the current work to other object detection and localization applications, such as medical image analysis, face detection, and so on.