This article discusses a novel approach to restoring astronomical images from Hubble Space Telescope (HST) quality to James Webb Space Telescope (JWST) quality through a Transformer model utilizing transfer learning. The study explores a structured method of generating datasets for training, where ground truth (GT) galaxy images were prepared, degraded to low-quality, and used for model finetuning. Testing demonstrated that the transformed images showed a significant improvement in correlation with their GT counterparts, indicating a breakthrough in image processing techniques despite certain limitations such as noise degradation and artifacts.
Our approach utilizes a Transformer model via transfer learning to restore astronomical images, enhancing them from HST quality to JWST quality.
The restoration process involved creating a pretraining dataset from GT galaxy images and degrading them to low quality to fine-tune our model.
Our results demonstrate that the restored images exhibit significantly improved correlations with ground truth images, showcasing the capabilities of our image restoration techniques.
Despite some limitations, our method presents a promising advancement in astronomical image processing, particularly in restoring images affected by noise and other artifacts.
Collection
[
|
...
]