Improving Colour Image Forgery Detection Using Machine Learning Methods

Abstract

Purpose: This study aims to develop an efficient and accurate approach for detecting copy-transfer image forgery, where duplicated image regions are geometrically transformed and inserted elsewhere in an image. The study aims to reduce computational complexity and detection time while maintaining high detection performance. Design/Methodology/Approach: The proposed method combines histogram-based features, Discrete Cosine Transform (DCT) features, and Iterative Closest Point (ICP)-based recurring-point features for copy-transfer forgery detection. The method extracts features from authentic and forged image data, then classifies them using a One-Class Support Vector Machine (OC-SVM). Particle Swarm Optimisation (PSO) optimises the classifier parameters. Research Limitation: The proposed approach is evaluated primarily on datasets containing copy-transfer forgeries. Performance may vary when applied to other types of image manipulations or under significantly different image acquisition and compression conditions. Findings: The proposed method successfully detected and localised copy-transfer forgery in the evaluated databases. The abstract reports an accuracy of 97.1%, precision of 94.1%, and recall of 95.5%. The reported execution times were 2.52 seconds and 3.07 seconds for the two evaluated databases. These results indicate that combining histogram-based information with DCT and ICP features can provide a practical balance between detection performance and computational cost. Practical Implication: The proposed model provides a computationally efficient solution for digital image authentication, making it suitable for applications such as digital forensics, media verification, law enforcement investigations, and content integrity assessment. Social Implication: By improving detection of manipulated digital pictures, the method helps combat misinformation, enhance trust in digital media, and support the authenticity of visual evidence in public and professional contexts. Originality/Value: This study shows a novel integration of histogram-based analysis, DCT features, and ICP spot recurring features combined with OC-SVM classification. The approach effectively addresses time-complexity challenges while maintaining high detection performance, offering a reliable and efficient solution for copy-transfer picture forgery detection.