Joint Deconvolution of Astronomical Images in the Presence of Poisson Noise

Axel Donath,Aneta Siemiginowska,Vinay L. Kashyap,David A. van Dyk,Douglas Burke

2024-03-21

Abstract:We present a new method for joint likelihood deconvolution (Jolideco) of a set of astronomical observations of the same sky region in the presence of Poisson noise. The observations may be obtained from different instruments with different resolution, and different point spread functions. Jolideco reconstructs a single flux image by optimizing the posterior distribution based on the joint Poisson likelihood of all observations under a patch-based image prior. The patch prior is parameterised via a Gaussian Mixture model which we train on high-signal-to-noise astronomical images, including data from the James Webb Telescope and the GLEAM radio survey. This prior favors correlation structures among the reconstructed pixel intensities that are characteristic of those observed in the training images. It is, however, not informative for the mean or scale of the reconstruction. By applying the method to simulated data we show that the combination of multiple observations and the patch-based prior leads to much improved reconstruction quality in many different source scenarios and signal to noise regimes. We demonstrate that with the patch prior Jolideco yields superior reconstruction quality relative to alternative standard methods such as the Richardson-Lucy method. We illustrate the results of Jolideco applied to example data from the Chandra X-ray Observatory and the Fermi-LAT Gamma-ray Space Telescope. By comparing the measured width of a counts based and the corresponding Jolideco flux profile of an X-ray filament in SNR 1E 0102.2-721} we find the deconvolved width of 0.58+- 0.02 arcsec to be consistent with the theoretical expectation derived from the known width of the PSF.

Instrumentation and Methods for Astrophysics,High Energy Astrophysical Phenomena

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to jointly deconvolve a set of astronomical observation images of the same sky area in the presence of Poisson noise. Specifically, the paper proposes a new method - Joint Likelihood Deconvolution (Jolideco) to reconstruct a single flux image by optimizing the posterior distribution based on the joint Poisson likelihood of all observations under multi - instrument observation conditions with different resolutions and point spread functions (PSF).

### Key problems

1. **Image degradation at low signal - to - noise ratio**:
- The quality of astronomical images is affected by the background, instrument measurement noise, and non - uniform exposure. Especially in the case of low signal - to - noise ratio, the images will be affected by Poisson noise.

2. **Observation differences of different instruments**:
- Different instruments have different resolutions and point spread functions (PSF), which lead to differences between the observation results. How to integrate these data from different sources to improve the reconstruction quality is an important problem.

3. **Limitations of existing methods**:
- Traditional deconvolution methods such as the Richardson - Lucy method (RL) are prone to decompose into multiple bright spots (speckles) when dealing with extended emission structures, and lack the estimation of the uncertainty of the reconstruction results.

### Solutions

The method Jolideco proposed in the paper solves the above problems in the following ways:

- **Joint Likelihood Deconvolution**: Optimize using the joint Poisson likelihood of all observation data instead of handling each observation separately.

- **Patch - based prior distribution**: Introduce a Gaussian Mixture Model (GMM) trained on high - signal - to - noise ratio astronomical images as a prior distribution. This prior distribution helps to capture the local correlation structures in the image without affecting the overall brightness or scale of the reconstructed image.

- **Improved reconstruction quality**: Verified by simulated data, Jolideco shows better reconstruction quality under multiple source scenarios and signal - to - noise ratio conditions, and is superior to the traditional Richardson - Lucy method.

### Experimental verification

The paper verifies by applying the Jolideco method to actual data (such as data from the Chandra X - ray Observatory and the Fermi - LAT gamma - ray space telescope), and shows its effectiveness in practical applications.

### Formula representation

The key formulas involved in the paper include:

- Poisson joint likelihood:
\[
P(D_j|\lambda_j)=\prod_{i = 1}^{N}\frac{e^{-\lambda_{j,i}}\lambda_{j,i}^{D_{j,i}}}{D_{j,i}!}
\]
where \(\lambda_j\) is the expected count after convolving the unknown flux image \(x\) by the point spread function (PSF) of a specific observation.

- Log - posterior distribution:
\[
L(x|D)=\sum_{j = 1}^{J}C(D_j|x)-\beta\cdot P(x)
\]
where \(C(D_j|x)\) represents the sum of the log - likelihoods of a single observation, \(P(x)\) is the log of the prior distribution, and \(\beta\) is a hyper - parameter for adjusting the prior weight.

Through these methods, Jolideco can effectively improve the reconstruction quality of astronomical images in the presence of Poisson noise.