This project implements a deep learning-based system to detect AI-generated images using forensic feature extraction techniques. The system combines CNNs with image forensics to achieve high accuracy in identifying synthetic content.
Team Members
- 126004237 - Saranya T S, ECE
- 126180060 - Baranika R, Electronics Engineering (VLSI)
- 126180019 - K Parvathavardhini Priya Sadhvi, Electronics Engineering (VLSI)
- 126180029 - Chethana Nagalli, Electronics Engineering (VLSI)
Demo Video Watch Demo on Google Drive
Project Files
- novelty.h5 – Trained model for AI image detection.
- ai_detector_gui.py – Tkinter GUI for real-time image authenticity verification.
- ai_img_detector.py – Feature extraction module.
Dataset Details
- For AI-Generated Images:(1219) https://www.kaggle.com/datasets/dibyarupdutta/dmimagesubset
- For Real Camera Captured Images:(1219)
Python Environment
- TensorFlow 2.17
- Keras 3.10
- NumPy 1.26
- Other dependencies: OpenCV, Pillow, scikit-learn, pandas.
Key Achievements
- Successfully developed a deep learning system to detect AI-generated images using features like ELA, PRNU, FFT, and noise residuals.
- Achieved 97.54% accuracy with the Fusion Advanced Model (CNN + ELA + PRNU + FFT + Noise Removal).
- Developed a user-friendly Tkinter GUI for real-time image authenticity verification.
Future Work
- Update models with newer AI-generated datasets (Midjourney, DALL-E 3, etc.).
- Explore advanced architectures: ResNet, Vision Transformers, EfficientNet.
- Apply data augmentation to increase dataset size and diversity for better training.
- Perform hyperparameter fine-tuning: learning rate, batch size, dropout, number of layers.
- Use Grad-CAM visualizations for model interpretability.