How itsakasht/sim_denoising works
Project type and purpose
• This is a scientific image-processing pipeline written in Python, focused on simulating Structured Illumination Microscopy (SIM), injecting noise, applying filters, reconstructing images, and evaluating filter performance.
• There is no client/server or UI. It’s a batch/offline data pipeline orchestrated via Python scripts and Jupyter notebooks.
File structure
• Top-level Python modules: noise.py, optics.py, denoising.ipynb, denoising_parts.ipynb, filter_evaluation.ipynb, frc_visualization.ipynb, color_bar.ipynb, visualization.ipynb.
• Supporting pipeline scripts (implied by README): denoising.py, denoising_parts.py, filter_evaluation.py, frc_visualization.py, color_bar.py (not present as .py, but code is in notebooks).
• Data directories:
• source_images (ground-truth TIFFs)
• output_images (subfolders by noise type & intensity, CSV filter results, final TIFFs)
• No network/config files, no database—pure file-based workflow.
Key architecture notes from README
• Modules:
• noise: adds Gaussian, Poisson, salt-pepper, mixed noise.
• optics: generates structured illumination, applies OTF, normalization, spatial/frequency display.
• denoising/denoising_parts: pipelines that apply illumination, add noise, apply filters, reconstruct SIM images. denoising_parts breaks data into patches for parameter tuning then stitches back.
• filter_evaluation: compares outputs from different filters, generates plots.
• frc_visualization: Fourier Ring Correlation analysis.
• color_bar: histogram/range visualizations.
Main components
a. Data sources
• “Ground Truth Images” (source_images folder)
b. Core processing modules
• Illumination Generator (optics module)
• Noise Adder (noise module)
• OTF Applier (optics)
• Filter Executor (denoising and denoising_parts)
• SIM Reconstructor (could be part of denoising scripts)
c. Evaluation & Visualization
• Filter Evaluation (filter_evaluation)
• Fourier Ring Correlation (frc_visualization)
d. Data sinks
• output_images folder (TIFFs, CSVs, plots)
Relationships and data flow
1) Start: Ground Truth Images → Illumination Generator
2) → produce 9 phase‐angle images → pass to Noise Adder
3) → noisy images → pass to OTF Applier → noisy+OTF images
4) → pass to Filter Executor
• denoising.py applies global filters
• denoising_parts.py applies patch‐wise filters + stitching
5) → output: reconstructed SIM images (noise vs. noise-free)
6) → pass outputs to Evaluation modules
• filter_evaluation compares metrics, plots PSNR, SSIM
• frc_visualization computes FRC curves
• color_bar generates histograms/range bars
7) → all results written to output_images/<noise_type>/<intensity>/
Architectural patterns and design principles
• Modular, single-responsibility modules: noise vs. optics vs. evaluation.
• Pipeline orchestration: Jupyter notebooks (denoising.ipynb, etc.) act as high-level workflows.
• File-based I/O: no in-memory service bus.
• Patch-wise vs. global processing (denoising_parts vs. denoising) as two interchangeable strategies.
Technologies and frameworks
• Python (core modules)
• Jupyter notebooks (workflows & visualization)
• NumPy/SciPy for arrays, FFTs
• Matplotlib/Seaborn for plots
• TIFF I/O (via pillow or tifffile)
• CSV generation for metrics