ML Image Denoising

A comprehensive suite of image denoising techniques, from classical statistical methods to state-of-the-art deep learning models. This repository provides implementations and comparisons of Principal Component Analysis (PCA), Non-negative Matrix Factorization (NMF), Singular Value Decomposition (SVD), Spectral Decomposition, Convolutional Neural Networks (CNN), and a Generative Adversarial Denoising Autoencoder (GAN-DAE).

-->

🎯 Overview

Image denoising is a fundamental problem in computer vision and image processing. This project serves as both a practical toolkit and an educational resource for understanding and comparing various denoising methodologies across traditional and modern approaches.

Key Features

• Classical Methods: Implementation of foundational techniques including PCA, NMF, SVD, and Spectral Decomposition
• Deep Learning: Custom Convolutional Neural Network designed for effective noise reduction
• State-of-the-Art: Sophisticated Generative Adversarial Network-based Denoising Autoencoder for high-fidelity reconstruction
• Comprehensive Evaluation: Built-in PSNR metrics and visual comparison tools
• Modular Design: Extensible codebase for easy experimentation with new techniques
• Interactive Notebooks: Jupyter notebooks for each method with detailed explanations

🚀 Quick Start

Prerequisites

• Python 3.6+
• Git
• pip
• virtualenv

Installation

1. Clone the repository

   git clone https://github.com/your-username/ml-image-denoising.git
   cd ml-image-denoising

2. Set up environment

   chmod +x install.sh
   ./install.sh

   This creates a virtual environment and installs all necessary dependencies.

3. Activate the virtual environment

   source bin/activate  # On Windows: Scripts\activate

Dataset Preparation

This project uses the CelebA dataset:

1. Download the dataset

   • Download img_align_celeba.zip from the CelebA project website
   • Extract to the parent directory: ../img_align_celeba/

2. Process the data

   python dataprocessing.py

   This script crops images to the required size and converts them to efficient HDF5 format.

📊 Usage

Jupyter Notebooks

Launch Jupyter to explore the different denoising methods:

jupyter notebook

Available Notebooks

• adversarial_autoencoder_denoising.ipynb - GAN-based denoising autoencoder
• cnn_denoising.ipynb - Convolutional Neural Network approach
• pca_denoising.ipynb - Principal Component Analysis method
• nmf_denoising.ipynb - Non-negative Matrix Factorization technique

Example Usage

# Load and preprocess data
from helper import cmp_images_psnr
import numpy as np

# Compare denoising methods
psnr_score = cmp_images_psnr(clean_images, denoised_images, num_samples)
print(f"PSNR: {psnr_score:.2f} dB")

🏗️ Project Structure

ml-image-denoising/
├── adversarial_autoencoder_denoising.ipynb  # GAN-DAE implementation
├── cnn_denoising.ipynb                      # CNN-based denoising
├── pca_denoising.ipynb                      # PCA method
├── nmf_denoising.ipynb                      # NMF technique
├── dataprocessing.py                        # Data preprocessing script
├── helper.py                                # Utility functions (PSNR, etc.)
├── install.sh                               # Environment setup script
├── requirements.txt                         # Python dependencies
└── lib/                                     # Core library modules
    ├── activations.py                       # Activation functions
    ├── updates.py                           # Optimization algorithms
    ├── ops.py                               # Neural network operations
    ├── metrics.py                           # Evaluation metrics
    └── ...

🔬 Methods Implemented

Classical Approaches

Method	Description	Notebook
PCA	Dimensionality reduction via principal components	pca_denoising.ipynb
NMF	Non-negative matrix factorization	nmf_denoising.ipynb
SVD	Singular value decomposition	pca_denoising.ipynb
Interpolation	Bilinear and bicubic upsampling	pca_denoising.ipynb

Deep Learning Approaches

Method	Description	Notebook
CNN	Custom convolutional neural network	cnn_denoising.ipynb
GAN-DAE	Generative adversarial denoising autoencoder	adversarial_autoencoder_denoising.ipynb

📈 Performance

Typical PSNR improvements on CelebA dataset:

• Bilinear Interpolation: ~25.5 dB
• Bicubic Interpolation: ~26.2 dB
• CNN Denoising: ~28-30 dB
• GAN-DAE: ~30-35 dB

Results may vary based on noise levels and training parameters.

🛠️ Configuration

Training Parameters

Key parameters can be modified in the notebook cells:

# Training configuration
batch_size = 128
learning_rate = 0.0002
num_epochs = 200
l2_regularization = 1e-5

Model Architecture

The GAN-DAE uses an encoder-decoder architecture:

• Encoder: Progressively downsamples from 128×128 to 8×8
• Decoder: Reconstructs high-resolution output
• Discriminator: Ensures realistic image generation

📋 Requirements

Core Dependencies

• Theano==0.7.0
• Pillow>=3.0.0
• numpy>=1.10.2
• h5py>=2.5.0
• tqdm>=3.4.0
• ipython>=4.0.0

See requirements.txt for complete dependency list.

🤝 Contributing

Contributions are welcome! Here's how you can help:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Areas for Contribution

• Implementation of additional denoising algorithms
• Performance optimizations
• Better evaluation metrics
• Documentation improvements
• Bug fixes and code quality improvements

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• CelebA Dataset for providing the facial image dataset
• Theano deep learning framework
• Original DCGAN implementation inspiration
• The open-source community for various utility functions

📞 Contact

For questions, suggestions, or collaboration opportunities:

• Issues: GitHub Issues
• Discussions: GitHub Discussions

🔗 Related Work

• Deep Image Prior
• DnCNN: Beyond a Gaussian Denoiser
• Noise2Noise: Learning Image Restoration without Clean Data

---

⭐ Star this repository if you find it helpful!