Image Forgery Detection Based on Convolutional Neural Network Using Hyperparameter Tuning

Image forgery detection is crucial because images are often manipulated, giving rise to misleading information that affects various areas, such as media, law enforcement, and science. There are several types of image forgery detection techniques, such as copy-move, retouching, and splicing, which are carried out for different types of image forgeries. High- resolution images contain more information, making it easier for forgers to create smooth and realistic edits. However, detecting small but significant changes in images with the use of traditional approaches is challenging. To overcome this problem, a Golden Sine Infinite Collapse based on the Walrus Optimization Algorithm with Convolutional Neural Network (GSI-WOA-CNN) is proposed for image forgery detection by applying hyperparameter tuning to the CNN for improved performance. The MICC-F220 dataset is a collection original and forged images, which are fed as input to the model. The collected images undergo pre-processing with techniques of image normalization and resizing, to improve the image quality. Hyperparameter tuning is performed using GSI-WOA, while forgery detection is classified using the CNN. The CNN achieves better performance accuracy compared to existing convolutional autoencoders. The performance of the proposed GSI-WOA-CNN model is evaluated in terms of accuracy, recall, specificity, F1-Score, False Positive Rate (FPR), and False Negative Rate (FNR). The evaluation results show performance values of 99.54%, 99.23%, 99.60%, 99.15%, 0.61, and 0.10, respectively, thereby proving the proposed GSI-WOA-CNN method's supremacy.