Literature Survey of Image Forgery Detection Using Machine Learning

Abstract

The rapid improvement in digital image manipulation means image forgery detection is more needed now than it ever was-be it artificially intelligently created deepfakes or adversarial altered images. Traditional techniques for error level analysis, copy-move detection, and splicing detection are too narrow to help forensics in keeping up with today's sophisticated AI-driven forgeries. The contribution of this survey is that it provides a detailed review of state-of-the-art machine learning-based forgery detection methods with particular emphasis on deep learning models, hybrid AI, and new emerging security technologies, such as Explainable AI, blockchain, and self-supervised learning. In general, approaches for image forgery detection may be classified as active or passive. This investigates the performance of state-of-the-art forgery detection techniques, ranging from feature-based machine learning classifiers (SVM, k-NN, and Decision Trees) to deep learning models, including CNNs and Vision Transformers. Yet, despite this progress, generalization, adversarial robustness, real-time efficiency, and dataset limitations are among the most well-known open issues that impede large-scale adoption. Among these, this investigate introduce the role of XAI methods such as Grad-CAM and SHAP in deep learning that can provide more transparent and interpretable deep learning models, an essential factor for forensic and legal applications. Going beyond deep learning, it explores blockchain-based image verification that allows decentralized and tamper-proof tracking of image authenticity. Besides, SSL and FSL are discussed as promising techniques that enable forgery detection models to learn new types of image manipulations with minimal labeled data. While deep learning models achieved impressive accuracy rates, their high computational demands raise challenges toward real-time deployment, thus opening space for research into lightweight AI models, federated learning, and edge computing solutions. Hybrid AI-driven forgery detection, robust deep learning models, and blockchain-backed verification systems will show the most promise for research in the time to come. In a nutshell, future research should go in the direction of making the AI detection systems more generalizable, efficient, and legally reliable against this growing threat to digital forgeries in media, forensics, and cybersecurity.