Multi-Resolution Data Fusion for Super Resolution Imaging

Applications in materials and biological imaging are limited by the ability to collect high-resolution data over large areas practical amounts of time. One solution this problem is low-resolution interpolate produce a image. However, most existing super-resolution algorithms designed for natural images, often require aligned pairing high training data, may not directly incorporate model sensor. In paper, we present Multi-resolution Data Fusion (MDF) algorithm accurate interpolation electron microscope at multiple resolutions up 8x. Our approach uses small quantities unpaired train neural network prior denoiser then Multi-Agent Consensus Equilibrium (MACE) formulation balance with forward agent that promotes fidelity measured data. xmlns:xlink="http://www.w3.org/1999/xlink">A key theoretical novelty analysis mismatched back-projectors, which modify typical updates computational efficiency or improved image quality. We use MACE prove using back-projector equivalent standard an appropriately modified model. microscopy results 4x 8x factors exhibit reduced artifacts relative methods while maintaining acquired accurately resolving sub-pixel-scale features.