An optimized CNN model for copy-move forgery detection and localization in digital images using particle swarm and Grey wolf optimization algorithms

Abstract

Image data is increasing daily at a rapid speed, and at the same time, image forgery has become widespread with readily available and accessible software tools. The image manipulations can be performed in different ways. But, copy-move image forgery is one of the prominent types of malicious forgery operation aimed at hiding sensitive information within the same image and making it challenging for human eyes to detect the forgery and authenticate its contents. The literature reported many deep learning techniques to detect the copy-move forgery images and enhance the detection accuracy. How ever, these techniques struggle with hyperparameter optimization issues, and fine-tuning hyperparameters and optimizing networks pose significant challenges. Hence, this study proposes an approach to implement optimized Convolutional Neu ral Network (CNN) model by integrating metaheuristic optimization strategies such as Particle Swarm Optimization (PSO) and Gray Wolf Optimization (GWO) algorithms to identify optimal CNN parameters for automated fine-tuning of CNN configurations. Further, the optimized CNN model is used to extract the significant image features and construct efficient feature maps to classify copy-move forgery images and accurately localize the forgery regions present within the images. To evaluate the performance of the proposed model, we used three publicly available standard datasets: CoMoFoD, CMFD, and MICC-F600. The CoMoFoD dataset is used to validate the performance against plain copy-move forgery images and forgery images with global postprocessing attacks such as image blurring, contrast reduction, noise addition, and JPEG compression. The CMFD and MICC-F600 datasets are used to test plain forgery images and forgery images with local postprocessing attacks such as rotation and scaling. The evaluation results show that the suggested approach effectively detects and localizes forgery, performing well against plain and post-processed copy-move forgery images.