Monday 10 March 2025
As technology continues to advance, our ability to manipulate images has become increasingly sophisticated. From photo editing software to deep learning algorithms, it’s now possible to alter images in ways that were previously unimaginable. But what happens when these manipulated images are used for nefarious purposes? That’s where image forensics comes in.
Image forensics is the process of detecting and analyzing altered or manipulated images. It’s a critical field that has real-world implications, from identifying deepfakes to detecting image tampering in legal proceedings. But until now, most forensic techniques have been limited by their inability to detect subtle changes in an image’s lighting and color balance.
That’s where Disharmony comes in – a new approach to image forensics that uses a combination of machine learning and physics-based models to detect even the slightest alterations to an image’s lighting and color. By training its algorithm on a dataset of real images and manipulated images, Disharmony is able to learn what normal lighting and color patterns look like, and then compare them to the patterns it detects in suspicious images.
One of the key benefits of Disharmony is its ability to detect subtle changes in an image’s lighting and color balance. Unlike other forensic techniques that rely on more obvious changes, such as object removal or addition, Disharmony can detect even the slightest alterations to an image’s tone and shading. This makes it a powerful tool for detecting deepfakes, which often involve subtle manipulations of an image’s lighting and color.
But how does it work? In short, Disharmony uses a combination of machine learning and physics-based models to analyze an image’s lighting and color balance. The algorithm is trained on a dataset of real images and manipulated images, and then uses this training data to learn what normal lighting and color patterns look like. It can then compare these patterns to the patterns it detects in suspicious images, allowing it to detect even the slightest alterations.
To test Disharmony’s abilities, researchers created a dataset of 40 unedited images and their corresponding edited versions using six different image harmonization methods. They then used Disharmony to analyze each image and compare its results to those produced by other forensic techniques. The results were impressive – Disharmony was able to detect even the slightest alterations to an image’s lighting and color balance, while other forensic techniques struggled to accurately identify edited regions.
The implications of Disharmony are significant.
Cite this article: “Detecting Deception: A New Approach to Image Forensics”, The Science Archive, 2025.
Image Forensics, Deepfakes, Machine Learning, Physics-Based Models, Lighting, Color Balance, Image Manipulation, Tampering Detection, Legal Proceedings, Artificial Intelligence
