Apppublic

yaxit24/Image-Forgery

sourceHugging Faceupdated 5mo agoView on Hugging Face

0likes

App README

🔍 Image Forgery Detection — Milestone 2

AI-powered forensic image analysis system combining deep learning (MobileNetV2) with classical ML (Random Forest) for forgery detection, localization, and structured reporting.

Features

—Dual-Model Classification: MobileNetV2 CNN + Random Forest ensemble
—Tampering Localization: Fused heatmap (ELA, PRNU residual, blockiness, copy-move) with Otsu thresholding
—Structured Reports: AI-generated forensic reports via Groq API (Llama 3.1)
—Export: PDF and Markdown report downloads
—Analysis History: Session-based tracking of analyzed images

Detectors

Method	What it catches
ELA (Error Level Analysis)	JPEG recompression inconsistencies
PRNU Residual	Sensor noise pattern anomalies
Blockiness Map	Block artifact grid inconsistencies
ORB Copy-Move	Duplicated regions via keypoint matching

Setup

bash

pip install -r requirements.txt
python train_cnn.py --data_dir data/train
streamlit run app/streamlit_app.py

Environment Variables

—GROQ_API_KEY: (Optional) Groq API key for AI-generated reports. Falls back to template-based reports if not set.