Enhanced Deep Learning-Based Feature Analysis for Copy-Move Forgery Detection

Abstract

Copy is a general operation method using a digital image, and a copy of the image is copied to the same image and inserted to hide or change the content. Tradition methods based on handmade characteristics (SIFT, SURF, ORB) aim at the geometric conversion and reliability of noise. This paper provides a deep structure for training to detect fake MAV copies using Sparkle Neural Networks (CNN) and the transformer-based model (VIT). This paper presents an enhanced deep learning-based feature extraction technique for CMFD using CNN architectures integrated with key point detection methods. Fake images are one of the most common types of images, and part of the image is replicated and placed in another location to mislead the audience. This fake detection is an important issue due to changes in lighting, texture, and geometric variations. In this paper, we have proposed an extended deep-trained method for copying fake detection. This method uses a spanning neural network (CNN) integrated with the caution mechanism to extract differential functions for forging images to ensure adaptability to actual scenarios. In addition, after processing, the description stage is introduced to increase localization accuracy using a method for comparing functions and converting evaluation. The experimental results show that the proposed model exceeds the existing modern methods in terms of accuracy, memory, and calculation efficiency. This research helps promote judicial image analysis and provides reliable and automated solutions to detect fake images in digital images. The proposed model combines the depth of signs, caution, and image segmentation to improve localization accuracy. The proposed approach reaches 98.2% detection accuracy, exceeding traditional and deep methods.