A multi-image super-resolution microscope system based on non-redundant information positioning

Abstract

Abstract This paper presents an advanced multi-image super-resolution (MISR) instrument to support higher-resolution measurement beyond the capability of the digital microscope, which is mainly contributed by the reconstruction of the subpixel images collected at the desired location. To address the challenges of low positioning accuracy and mismatched training examples encountered with microscopic systems, we developed tailored subpixel positioning and image training methods. Specifically, to capture images at the desired location, an improved image-positioning method based on non-redundant information of the subpixel images was proposed. To match the training model with the targets for MISR, we propose a novel method that constructs training examples by progressively fusing subpixel images captured at varying resolutions. Compared with existing methods, the proposed algorithm demonstrated 23.6% improvement and 37.1% improvement in PSNR and SSIM respectively.