Paper Title
A Novel Multi-Model Framework for Digital Image Forgery Detection

Abstract
With the rise of digital image modification, there is an increasing need for effective image forensics procedures. Deep learning algorithms like ResNet50, VGG16, XceptionNet, and MobileNetV2 are known to achieve great results in detecting falsified images. This research aims to assess the accuracy of the four recognized models and an ensemble model that combines ResNet50, VGG16, XceptionNet, and MobileNetV2 in performing image forgery detection tasks. For feature representation enhancement, the proposed approach employs Thepade Sorted Block Truncation Coding (TSBTC) for feature extraction and Error Level Analysis (ELA) for revealing parts of the image that have been altered. By employing an ensemble approach, we noticed measuring accuracy below 93.75, with the greatest losses incurred when integrating multiple models. Onthe contrary, independent models, especially ResNet50 and XceptionNet, provide better scores - measuring above 98.75. This research analyzes several possibilities that may contribute to an ensemble approach performing better than separate approaches, focusing on feature fusion, overall system design, and model parameterization. These findings are useful in improving the current methods of model image fusion so that images that are tampered with can be detected efficiently. Keywords: Image Forgery Detection, Deep Learning, TSBTC, ELA, CNN, Ensemble Learning, Resnet50, VGG16, Xceptionnet, Mobilenetv2, And CASIA V2.