Authors:
(1) Hyosun park, Department of Astronomy, Yonsei University, Seoul, Republic of Korea;
(2) Yongsik Jo, Artificial Intelligence Graduate School, UNIST, Ulsan, Republic of Korea;
(3) Seokun Kang, Artificial Intelligence Graduate School, UNIST, Ulsan, Republic of Korea;
(4) Taehwan Kim, Artificial Intelligence Graduate School, UNIST, Ulsan, Republic of Korea;
(5) M. James Jee, Department of Astronomy, Yonsei University, Seoul, Republic of Korea and Department of Physics and Astronomy, University of California, Davis, CA, USA.
Table of Links
2 Method
2.1. Overview and 2.2. Encoder-Decoder Architecture
2.3. Transformers for Image Restoration
4 JWST Test Dataset Results and 4.1. PSNR and SSIM
4.3. Restoration of Morphological Parameters
4.4. Restoration of Photometric Parameters
5.2. Restoration of Multi-epoch HST Images and Comparison with Multi-epoch JWST Images
6 Limitations
6.1. Degradation in Restoration Quality Due to High Noise Level
6.2. Point Source Recovery Test
6.3. Artifacts Due to Pixel Correlation
7 Conclusions and Acknowledgements
Appendix: A. Image restoration test with Blank Noise-Only Images
4.4. Restoration of Photometric Parameters
One of the immediate scientific utilities of image restoration is enhancing photometry. Here we compare aperture flux, isophotal flux, and individual pixel values measured from the RS images with the GT images to evaluate the performance in the photometric context.
Since we use min-max normalization consistently across our training and input datasets, the dynamic range of the RS images is also restricted. To extract photometry from the LQ and GT images, we opt to use the original (pre-normalization) images. Consequently, it is necessary to rescale the RS images. To ensure a fair comparison, this rescaling process must be executed independently, without relying on the information available from the corresponding GT images.
We aligned the dynamic range of the RS images to the LQ images as follows. First, we measured the lower
Figure 7 displays the resulting flux comparisons. The aperture fluxes from the RS images are in good agreement with those from the GT images. Compared to the LQ images, the scatter is reduced by ∼60%. The scatter reduction is similar (∼55%) in isophotal flux. It is
![Figure 7. Comparison of photometric information between GT and LQ images (blue) and between GT and RS images (red). We investigate correlations of aperture flux, isophotal flux, and individual pixel values. For flux comparison, since the RS images were scaled to the range [0, 1], we rescaled the RS images to enable quantitative comparisons (see text for details). We stress that we do not use the information from GT for rescaling. We use an elliptical aperture defined with SExtractor’s semi-major, semi-minor axes, and orientation angle from the LQ image. The isophotal area is also determined from the LQ image. For comparison of individual pixels, we use the pixels only within the elliptical aperture. Flattened pixel values in LQ images due to convolution are restored to their original values, reproducing a one-to-one slope. The photometric information is recovered remarkably well, with a significant reduction of scatter compared to the LQ-GT comparison.](https://hackernoon.imgix.net/images/fWZa4tUiBGemnqQfBGgCPf9594N2-xia315k.png?auto=format&fit=max&w=3840)
worth noting that the isophotal fluxes in the LQ images are systematically overestimated because the isophotal area is defined from the LQ image[5]. This bias is significantly reduced in the RS images. Finally, the pixel-to-pixel comparison illustrates a tight 1:1 correlation between the RS and GT images across the entire dynamic range, while the LQ images show a slope significantly less than unity because of their larger PSF. The pixel-to-pixel scatter reduction is by a factor of 7.
This paper is available on arxiv under CC BY 4.0 Deed license.
[5] That is, noise can make some pixel values near the edge of the isophotal area in the LQ image higher than the GT values.
