Computer Vision · Machine Learning · Digital ForensicsBuilt a feasibility proof of concept
AI-Manipulated Image Detection — PoC
An experimental machine-learning approach to assess whether document images may have been manipulated using AI.
Status
Pilot / experimental proof of concept
The Problem
AI-generated or manipulated images can make document review more difficult. The proof of concept explored whether handcrafted image-forensics signals could help identify suspicious images.
Solution & Contribution
Designed 56 handcrafted image-forensics features and trained a Logistic Regression classifier. The pilot achieved approximately 95% mean accuracy in 5-fold cross-validation on the pilot dataset.
My Contribution
Feature design, classifier training and cross-validation evaluation.
Technologies Used
PythonImage ProcessingHandcrafted FeaturesLogistic Regressionscikit-learnCross-Validation
Interested in collaborating?
Get in touch with Khalid to discuss applied AI engineering, LLMs, or agent systems.
kh.regui@gmail.com