PnP-Proximal-Graphical-Model: Plug-and-Play Proximal Graphical Model for Microscopic Image Demosaicing

PnP-Proximal-Graphical-Model (PnP-PGM) is a novel optimization-based framework for demosaicing microscopic vine wood images.
It leverages the Plug-and-Play (PnP) priors with a Proximal Gradient Model (PGM), integrating powerful denoisers such as BM3D to achieve superior color fidelity, structural preservation, and reduced artifacts compared to existing methods.

📄 Reference: Research Paper

✨ Key Features

• 🔹 PnP-PGM Framework – combines proximal optimization with plug-and-play denoisers.
• 🔹 BM3D Prior Integration – preserves fine textures and removes noise effectively.
• 🔹 Microscopic Vine Wood Dataset – 1324 fluorescence microscopy images for benchmarking.
• 🔹 Quantitative & Qualitative Evaluation – against PnP-ADMM and PnP-ADMM Consensus.
• 🔹 Superior Performance – achieves higher PSNR and SSIM, while reducing artifacts.

🏗️ Method Overview

🔹 Proposed PnP-PGM Framework

🔹 Workflow

1. Raw mosaiced image acquisition.
2. Proximal gradient descent update for fidelity.
3. BM3D denoising as a PnP proximal operator.
4. Iterative refinement until convergence.
5. Post-processing for hue and structural correction.

📊 Results

Quantitative Comparison

Method	PSNR (dB)	SSIM
PnP-ADMM	40.76	0.9053
PnP-ADMM Consensus	40.83	0.9032
Proposed PnP-PGM	41.06	0.9095

✅ PnP-PGM achieves the highest PSNR and SSIM, with improved texture preservation and reduced color artifacts.

📂 Dataset

• Microscopic Vine Wood Dataset
  • 1324 fluorescence microscopy images.
  • Designed for pathogen segmentation and demosaicing research.
• Public Sources:
  • UCI Machine Learning Repository
  • HAL Archive – Upper-Rhine AI Symposium 2024

⚙️ Implementation Details

• Framework: Python 3.6
• Denoiser: BM3D Plug-and-Play Prior
• Optimization: Proximal Gradient Descent
• Parameters:
  • Step size α ∈ [0.68, 0.74]
  • Denoising strength σ ∈ [0.028, 0.036]
• Evaluation Metrics: PSNR, SSIM, Visual Quality

🚀 Getting Started

Installation

git clone https://github.com/OVER-CODER/PnP-Proximal-Graphical-Model.git
cd PnP-Proximal-Graphical-Model
pip install -r requirements.txt

Run the Algorithm

python PnP_PGM.py --input Dataset/sample_image.png --output Results/output.png

📌 Citation

If you use PnP-PGM in your research, please cite:

@article{pandit2025demosaicing,
  title={Demosaicing of Microscopic Vine Wood Images Using Plug-and-Play Proximal Graphical Models},
  author={Pandit, Aryan and Kumar, Anurodh and Vishwakarma, Amit},
  year={2025},
  journal={Journal of LaTeX Class Files}
}

🤝 Acknowledgments

• BM3D developers for the denoiser integration.
• Dataset contributors [22, 23].