Enhancing Astronomical Image Stitching with XFeat: A Deep Learning Approach

Image stitching is a widely applicable technique, with its importance magnified in the realm of astronomical research, where the assembly of high-quality composite images is essential for the accurate measurement and detailed analysis of celestial bodies. The limitations of astronomical cameras, including their restricted fields of view and the propensity for wide-angle lenses to introduce fisheye distortions, present significant challenges in capturing large field-of-view astronomical images in a single exposure. Consequently, the stitching of multiple images emerges as a critical technique for the acquisition of expansive astronomical imagery. In this paper, a novel deep learning-based approach to astronomical image stitching is introduced. The method leverages deep neural networks to automatically extract keypoints from images and perform feature matching, thereby significantly outperforming traditional stitching techniques. Termed XFeat, the proposed method harnesses the power of convolutional neural networks (CNN) to identify and align features across images with high precision, resulting in a substantial enhancement in the quality of stitched astronomical images.