YourLyingEyes is an advanced image forensics tool that leverages a combination of commercial and non-commercial AI models and attempts to determine the authenticity of digital images. Python must be installed.
1. Error Level Analysis (ELA): This technique detects image manipulation by analyzing compression artifacts. By saving the image at a known compression level and comparing it to the original, ELA highlights discrepancies in compression levels, revealing potential edits. It is particularly effective for JPEG images and visualizes tampering as a heatmap. (https://github.com/jayant1211/Image-Tampering-Detection-using-ELA-and-Metadata-Analysis)
2. ImageForensics Modelling of Predictable and Unseen Noise Introduced by Online Social Networks (OSNs): This model enhances forgery detection by analyzing and decoupling noise introduced by OSNs into predictable and unseen noise. It incorporates these noises into a robust training framework, highlighting discrepancies in compression artifacts to indicate potential tampering. (https://github.com/highwaywu/imageforensicsosn)
3. Digital Forensics Net: A deep neural network designed for pixel-wise image forgery detection. It excels in identifying manipulated regions within images, even after common lossy operations like resizing and compression. The model generates an output image where each pixel is classified as either authentic or forged, visually highlighting tampered areas. It is for non-commercial use. (https://huggingface.co/spaces/DFisch/Image-Manipulation-Detection/tree/main)
4. Compression Artifact Tracing Network (CAT-Net): An end-to-end fully convolutional neural network designed to detect and localize image splicing. It leverages both RGB and DCT streams to learn forensic features of compression artifacts, distinguishing between authentic and tampered regions. The network outputs a probability map for each pixel, indicating the likelihood of manipulation. It is for non-commercial use. (https://github.com/mjkwon2021/CAT-Net)
5. Multi-Modal Fusion for Image Manipulation Detection: Auxiliary representations from forensic filters are passed through a convolutional block to produce early features, which are concatenated and further processed to create mixed features. These mixed features, along with the original RGB image, are used as input for a dual-branch CMX encoder. This approach leverages the complementary capabilities of different forensic filters, resulting in a robust and stable optimization for detecting image manipulations. It is for non-commercial use. (https://github.com/idt-iti/mmfusion-iml)
