Image Forgery Detection
Deep-learning system to detect and localize manipulated regions in images (Copy-Move, Splicing).
Overview
This project automates detection of manipulated images using a Convolutional Neural Network (CNN). It performs preprocessing (resizing, normalization, noise analysis), trains a CNN classifier and produces both a binary authenticity decision and a heatmap overlay that highlights suspicious regions. Use cases include digital forensics, media verification and evidence validation. (Details & performance are taken from the project report.)
Problem
Images can be tampered by copy-move, splicing or advanced editing. Manual inspection is slow and error-prone.
Solution
- Preprocess images (noise, EXIF checks, normalization).
- CNN model trained on labeled dataset (original vs forged) to detect tampering.
- Generate heatmap overlay for localization and visualization.
Architecture & Tech
Model (summary)
Typical CNN stack — Conv → Pool → Conv → Pool → Flatten → Dense → Dropout → Softmax for binary classification. Training includes augmentation (rotate/scale), and evaluation uses precision/recall/F1. Sample code and plots were included in the project report.
Tech Stack
Python, TensorFlow / Keras, OpenCV, NumPy, Matplotlib, Flask (demo web), and optional deployment on cloud (GCP/AWS) for scalable inference.
Results & Demo
Key results reported in the report: >90% accuracy on a test split, clear heatmap localization for splicing/copy-move cases. Users can upload images in the web demo and view overlays.
Project Highlights
- Achieved high detection accuracy with a robust preprocessing pipeline.
- Integrated visualization overlays (heatmaps) for explainability — useful in forensic workflows.
- Implemented a simple Flask demo allowing image upload and real-time inference.
Python — Online Modules
Forage — Software Engineering Simulation (Mar 22, 2025)
SQL — SQL using AI workshop
The Website Makers — Java Programming