Deep Learning-Based Image Forgery Detection Using CNN and UNet for Precise Tampered Region Identification

Abstract: This research focuses on detecting forged images using a Convolutional Neural Network (CNN) for classification and a Dual-Stream UNet (D-UNet) for localizing manipulated regions. The system leverages Error Level Analysis (ELA) and Spatial Rich Model (SRM) filters to enhance forgery detection accuracy. The proposed approach provides a probability score for authenticity and highlights tampered areas, ensuring a robust and interpretable forgery detection framework With the increasing accessibility of digital image editing tools, image forgery has become a significant concern in various fields, including journalism, forensics, and security. This paper presents an advanced approach to detecting image forgery using deep learning techniques, particularly Convolutional Neural Networks (CNNs). Our method incorporates both traditional forgery detection techniques such as Error Level Analysis (ELA) and Frequency Domain Analysis, along with a dual-stream U-Net model. The first stream processes raw RGB images, while the second stream analyzes filtered images using Spatial Rich Model (SRM) features to capture subtle inconsistencies introduced during forgery. The combined feature representations are then used for classification, distinguishing between authentic and tampered images. Experimental results on benchmark datasets, including CASIA and Co Mo Fo D, demonstrate that our approach outperforms existing methods in terms of accuracy, precision, and recall. The proposed method not only enhances forgery detection capabilities but also contributes to the ongoing efforts in ensuring digital image integrity.

Keywords: Image Forgery Detection, Convolutional Neural Networks, U-Net, Error Level Analysis, Spatial Rich Model, Digital Forensics.