I Built TRACE-X: Exploring Digital Image Forensics Beyond Basic Tampering Detection
Images are shared across social media, messaging platforms, and websites every day.
But what happens to forensic evidence when an image gets compressed, resized, cropped, or edited before being shared?
That's the question that led me to build TRACE-X 2.0 — Transformation-Resilient Adaptive Cross-platform Evidence Engine.
The Problem
Traditional image-forensics techniques can become less reliable when images undergo common transformations.
A forensic signal that is visible in an original image might weaken or change after JPEG compression, resizing, or cropping.
I wanted to explore whether combining multiple forensic signals could provide more useful evidence under these conditions.
What is TRACE-X?
TRACE-X is an experimental digital image-forensics framework that combines multiple image-analysis techniques into a single evidence-fusion pipeline.
Currently, it explores four forensic signals:
Error Level Analysis (ELA): Examines differences introduced by JPEG recompression.
Discrete Cosine Transform (DCT): Analyses frequency-domain characteristics.
Noise Residual Analysis: Examines residual patterns after image smoothing.
JPEG Grid Analysis: Examines discontinuities around JPEG block boundaries.
These signals are combined using a heuristic, profile-conditioned fusion mechanism.
The system generates:
Combined evidence maps.
Evidence overlays.
Exploratory threshold masks.
Signal disagreement visualisations.
Machine-readable JSON reports.
Human-readable HTML reports.
Testing Transformation Robustness
I also implemented experiments involving:
JPEG compression at different quality levels.
Image resizing.
Cropping.
Gaussian blur.
The objective is to understand how different forensic signals behave when the same image undergoes these transformations.
Tech Stack
Python
NumPy
OpenCV
Pillow
SciPy
Matplotlib
Streamlit
Pytest
What's Next?
TRACE-X is still an experimental research project.
My next goals are:
Integrate established image-forensics datasets.
Evaluate localisation using ground-truth manipulation masks.
Compare against existing forensic baselines.
Explore learned transformation profiling.
Develop and validate adaptive evidence fusion.
Investigate cross-dataset generalisation.
I am particularly interested in understanding whether transformation-aware evidence fusion can improve the localisation of manipulations compared with fixed forensic approaches.
Explore the Project
GitHub: https://github.com/sreevatsarajesh/Trace-X
I welcome feedback, suggestions, research references, and contributions from anyone working in computer vision, cybersecurity, or digital forensics.
If you've worked on image manipulation detection or forensic localisation, I'd love to hear your thoughts.
Building, testing, learning, and improving. 🚀
Top comments (0)