A Novel Hybrid Model to Detect Image Tempering

Abstract

With the growing use of social media and mobile apps in daily life, the ability to alter digital images has significantly increased. In fact, because of digitalization, images are often considered more trustworthy than words, yet digital image forgery has become one of the most recognized issues for people who regularly use social media and apps. The availability of affordable mobile phones and other electronic devices, along with various applications, has made it easy to capture, store, and share images on social media, making them very common. Moreover, the presence of user-friendly software editing tools allows even those with little or no technical experience to modify or alter images. There is no longer a need for advanced skills in creating forgeries or manipulating digital images, which has led to a greater risk of compromising the authenticity and integrity of images due to technological progress. In the past, such tasks required specialized knowledge, but with the rapid development of sophisticated editing tools and software, altering or forging digital images has become much simpler. Furthermore, detecting altered images with the naked eye is now very difficult, and in many cases, almost impossible, especially when the forgery is done skillfully. There are often no visible signs of tampering. As a result, digital images in media are no longer reliable, and image tampering has become more common. Therefore, developing algorithms to verify the authenticity of digital images has become essential, particularly in cases where images are used as evidence in court, financial, or medical contexts. Hence, detecting digital image forgery has become a major focus of digital image forensics and is also important in everyday use, as without the original image, it is challenging to identify any signs of forgery. Additionally, when part of an image is copied and pasted into another part of the same image—either unchanged or with some transformation—it becomes very difficult to detect the altered sections, especially since the copied regions can closely resemble the original. For these reasons, the need for digital forgery detection remains critical as outlined above.This thesis presents a detailed hybrid framework designed to identify tampered areas in digital images.

Tampering, such as adding, removing, cloning, or making minor changes to objects in an image, poses serious threats to the credibility of visual media. This is particularly concerning in fields like journalism, legal documentation, medical imaging, and national security. The key innovation of this method lies in its comprehensive approach, which effectively combines three traditional statistical methods—Error Level Analysis (ELA), Noise Residual Estimation, and Copy-Move Forgery Detection—into a single, format-agnostic forensic solution. Unlike many existing methods that rely on specific formats or are limited to certain types of tampering, this approach offers a universal hybrid technique supported by adaptive thresholding and intuitive red-shadow visual markers. A user-friendly MATLAB GUI has been developed to enable investigators across various fields to use this system effectively.