Authors:
(1) Hyosun Park, Department of Astronomy at the University of Yonsii University of Korea;
(2) Yongski Jo, an artificial intelligence school, sleep, Ulsan, Republic of Korea;
(3) Seokun Kang, School of Artificial Intelligence, Sleep, Ulsan, Republic of Korea;
(4) Taehwan Kim, a school that graduated from artificial intelligence, sleep, Ulsan, Republic of Korea;
(5) M. James Jee, Department of Astronomy at the University of Yonsii, Soul, Republic of Korea and Department of Physics and Astronomy, California University, Davis, Ca, USA.
Table
Summary and 1 Introduction
2 method
2.1. Overview and 2.2. Architecture of the Coder-Deplifier
2.3. Transformers to restore the image
2.4. Details of implementation
3 data and 3.1. HST data
3.2. Galsim data
3.3. Jwst data
4 JW test data results and 4.1. PSNR and SSIM
4.2. Visual inspection
4.3. Restoring morphological parameters
4.4. Restoring photometric parameters
5 Application for real HST pictures and 5.1. Restoring and comparison with diverse pictures of one-e-success
5.2. Restoring and comparison with multilateral jwst images
6 restrictions
6.1. The decomposition of recovery quality due to high noise level
6.2. Test for the point of the point sources
6.3. Artefacts of pixel correlation
7 conclusions and recognitions
Appendix: A. Image recovery test with empty noise pictures
References
3.2. Galsim data
To generate pre-training data, we use Galsim's HST DATA DATESTS Based on the distribution of serial parameters, to reproduce not only the marginalized distribution of each parameter, but also correlations between the parameters. We constructed the probability functions (PDF) in the five -dimensional parameter space (axis ratio, serial index, flow, size and noise) with scipy.Tats Gaussi Kernel density assessor (KDE) module. Thereafter, 100,000 new Galsim input parameter sets were generated on the basis of derived PDFs.
Image size W varies in the original HST postage damp data. However, the galaxy images in the training data must be of equal size. We choose the size of 64 × 64 pixels, as the choice gives a good balance between efficiency and the loyalty of the image. The smaller image cuts the profile in the league prematurely, while the larger picture is dominated by background values.
The 100,000 noise-free galaxy images issued by Galsim form GT images in the preservation data. Thereafter, these GT images had come into contact with a 7 × 7 pixel Gauss nucleus, whose width corresponds to the HST PSF. Finally, the noise of Gauss was added to bring the RMS noise level with HST. These deteriorated images consist of LQ images.
Following the same procedure used to determine the most appropriate serial parameters of HST galaxy images, we measured the Serzic parameters of LQ images. Figure 4 compares the Sersic parameter distribution between HST and Galsim LQ images. It must be understood that, in principle, two divisions cannot be exactly because the input used to generate Galsim image -serial parameters are derived from noisy HST images. However, we believe that in general, covarations and the chances of excluded probabilities are similar to the two data sets. Because Galsim galaxy images are used to pre-form a model, which JWS-based images will end later, we do not think that differences in detail are important.
3.3. Jwst data
We used almost the same procedures used to create HST data to try the JWS image. The only difference is the noise level criteria. As the JWst images act as a GT before decay, the noise level must be much lower. We only accepted the picture if its RMS noise was less than 0.02 after the min-maxi normalization.
One non -trivial question is how to deal with the difference in pixel scale between JW and HST. The natural pixel scale of the Nircami is 0.03 ′ ′ (0.06 ′ ′) for a short (long) wavelength channel that differs from the HST/ACS pixel scale 0.05 ′ ′. We weighed the JWs pixel scale to re -try the value of 0.05 ′ ′. However, interpolation noise was impaired by the original quality. Therefore, we kept the original scale. Therefore, our JWS galaxies are 40% larger or 20% smaller than the HST average. However, as we demonstrate in §4, this pixel scale is not significantly affected by the restoration of our HST images.
Another complex topic is JW's inner PSF. Although the JWS PSF is on average much smaller than the HST, its long wavelength channel PSF width may be similar to that of HST/ACS PSF. You can consider deconvolating the image of JWS with your PSF as an attempt to make your resolution even better and thus better represent the truth. It may -that is a achievable approach that resembles the current method studied here. But it is outside the scope of this study. In the current study, we use JWs without a deconvolution. Thus, the Effective PSF of LQs described below includes both JW and HST PSF conventions.
We standardize the size of the post-stamp image to 64 × 64 pixels. The smaller image is extended to its four edges by padding with pixels so that both the noise and the background level correspond to the original image. The larger picture is cut to 64 × 64 pixels. We abandoned the galaxy of pictures whose profiles were prematurely trimmed by the cutting procedure. We collected JWst Nircam F115W, F200W, F2777W and F444W images of various areas available publicly available, publicly available in August 2023 in August 2023, Mikaski Archives for Space Telescopes (Mast) Image (Mast) Image (Mast)[4]To. Dataset is a GT component in our Finetuning data.
LQ -Generates JWS GT Galaxy when concentrating pictures with HST PSF and adding Gaussian noise. Although the pixel scale differs from the HST picture, we scaled the size of the HST PSF, assuming the 0.05 ′ ′ pixel scale. This is due to the fact that we finally want to restore HST images with a pixel scale of 0.05 ′. The applied noise distribution follows the HST data. The relationship between training, validation and test data sets is 8: 1: 1.
[4]
