Arbitrary scale image super-resolution reconstruction method

Reconstructing images through super-resolution techniques plays a critical role in digital image processing research. The objective is to leverage details from low-resolution images to generate a high-resolution counterpart. With the advancement of computer technology, the theoretical value and practical significance of image super-resolution reconstruction have become increasingly prominent, making it a key area of interest within computer vision research. The goal of super-resolution reconstruction for arbitrary-scale images is to recover high-quality, high-resolution images from low-resolution images without limiting magnification. This approach provides flexibility, allowing for any amplification factor to be processed, rather than being constrained to a predefined number of amplification levels. In this way, accurate reconstruction of fine image details can be achieved, regardless of the magnification scale, resulting in clear and detailed high-resolution images. To achieve super-resolution at any scale, researchers have developed new modules that can dynamically generate filters based on the input scale information, allowing the neural network to adapt to a range of different scaling factors, thereby enabling super-resolution reconstruction at arbitrary scales. Among these, LIIF (Local Implicit Image Function) and LTE (Local Texture Estimator) are two notable contributions to arbitrary-scale image super-resolution reconstruction. This article focuses on these two works and optimizes the existing method structure.