Loknaath/Image_forgery_detection

🖼️ Image Forgery Detection using ELA + CNN

Welcome to the Image Forgery Detection demo! This Space provides a real-time web application that uses Error Level Analysis (ELA) 🔍 and a Convolutional Neural Network (CNN) 🤖 to determine whether an image is Authentic or Forged.

Upload an image and the system will:

1. 1.🔧 Convert it to its ELA representation
2. 2.🧠 Analyze it using a trained EfficientNetB0 model
3. 3.🟢 Predict Authentic or 🔴 Forged with confidence
4. 4.🖥️ Display the ELA visualization

🔬 How It Works

1️⃣ Error Level Analysis (ELA)

ELA highlights areas with inconsistent compression. Edited or manipulated regions often show brighter or uneven artifacts in the ELA output. ✨

2️⃣ CNN Classification

A deep-learning model (EfficientNetB0) trained on the CASIA 2.0 dataset evaluates the ELA-transformed image and performs binary classification:

• —✅ Authentic
• —❌ Forged

📂 Dataset

CASIA 2.0 Image Tampering Detection Dataset

Contains both authentic and forged images involving:

• —✂️ Splicing
• —🔁 Copy-move
• —🧽 Object removal
• —🖼️ Image composition

Dataset was preprocessed using ELA and split into:

• —80% 🏋️ Training
• —20% 🧪 Validation

🛠️ Technologies Used

• —🐍 Python
• —🧠 TensorFlow / Keras
• —⚙️ EfficientNetB0
• —🖼️ PIL (for image processing & ELA)
• —🌐 Gradio (UI)
• —☁️ Hugging Face Spaces

👥 Project Members

• —👤 Loknaath P – Reg No: 212223240080
• —👤 Lokhnath J – Reg No: 212223240079

🌍 Application Areas

• —🕵️ Digital forensics
• —📰 Fake media detection
• —🔍 Journalism & fact-checking
• —⚖️ Legal investigation support
• —🎓 Academic integrity

🎯 Project Objective

Provide a simple, accessible online tool that helps detect digitally manipulated images using forensic analysis and deep learning.

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference