A Zero-Shot Super-Resolution Image Reconstruction Technique Based on Radial Basis Function Neural Networks

This paper addresses the challenge of improving image resolution in special fields, such as medicine and astronomy, where high-quality and high-resolution image sample data is difficult to obtain. We propose a novel zero-shot super-resolution image reconstruction method that leverages machine learning to obtain the mapping relationship between the original image and its degraded image. We can improve image resolution by learning the similar structures present in images across different scales and angles. Our method consists of three processes, namely expansion, deblurring, and edge enhancement, all of which use radial basis function neural networks to learn the mapping relationship between the original image and its degraded image. Our comparative experiments on the Set5 dataset show that our method's performance is comparable to that of the bicubic interpolation algorithm. Moreover, we can further improve image resolution by using our method directly on top of the bicubic interpolation algorithm, and our experimental results demonstrate that our method outperforms the bicubic interpolation algorithm. Importantly, our method is portable and can be combined with other super-resolution algorithms to enhance image resolution further.