PyWI (Python wavelet imaging)

Introduction

PYWI is an image filtering library aimed at removing additive background noise from raster graphics images.

• Input: a FITS file containing the raster graphics to clean (i.e. an image defined as a classic rectangular lattice of square pixels).
• Output: a FITS file containing the cleaned raster graphics.

Description

The image filter relies on multiresolution analysis methods (Wavelet transforms) that remove some scales (frequencies) locally in space. These methods are particularly efficient when signal and noise are located at different scales (or frequencies). Optional features improve the SNR ratio when the (clean) signal constitute a single cluster of pixels on the image (e.g. electromagnetic showers produced with Imaging Atmospheric Cherenkov Telescopes). This library is written in Python and is based on the existing Cosmostat tools iSAp (Interactive Sparse Astronomical data analysis Packages).

The PyWI library also contains a dedicated package to optimize the image filter parameters for a given set of images (i.e. to adapt the filter to a specific problem). From a given training set of images (containing pairs of noised and clean images) and a given performance estimator (a function that assess the image filter parameters comparing the cleaned image to the actual clean image), the optimizer can determine the optimal filtering level for each scale.

The PyWI library contains:

• wavelet transform and wavelet filtering functions for image multiresolution analysis and filtering;
• additional filter to remove some image components (non-significant pixels clusters);
• a set of generic filtering performance estimators (MSE, NRMSE, SSIM, PSNR, image moments difference), some relying on the scikit-image Python library (supplementary estimators can be easily added to meet particular needs);
• a graphical user interface to visualize the filtering process in the wavelet transformed space;
• an Evolution Strategies (ES) algorithm known in the mathematical optimization community for its good convergence rate on generic derivative-free continuous global optimization problems (Beyer, H. G. (2013) “The theory of evolution strategies”, Springer Science & Business Media);
• additional tools to manage and monitor the parameter optimization.

Dependencies

• Python 3.5 or 3.6
• Numpy
• Scipy
• Scikit-image
• Pillow (a.k.a. PIL) to read and write many image formats (PNG, JPEG, TIFF, ...)
• Astropy to provide Fits file format
• Matplotlib 1.5 or later to provide plotting functionality
• Pandas
• Cosmostat iSAP Sparce2D

Installation

1. Download and install the iSAP library
2. Install the PyWI library
   In your system terminal, type: pip install pywi
3. Test

Download a sample image (e.g. archives_ngc3576.png)

In your system terminal, type:

pywi-mrtransform -t 256,256,256,0 --plot archives_ngc3576.png

pywi-mrfilter -s 256,256,256,0 --plot archives_ngc3576.png

4. Type pywi-mrtransform -h or pywi-mrfilter -h to display the list of available options and their documentation.

More details on the installation procedure are available on the full documentation page.

Authors and Contact

• Jérémie Decock
• Tino Michael
• Thierry Stolarczyk
• Karl Kosack