owards adaptable synchrotron image restoration pipeline

br />Synchrotron microscopic data commonly suffer from poor image quality with degraded resolution incurred by instrumentation defects or experimental conditions. Image restoration methods are often applied to recover the reduced resolution, providing improved image details that can greatly facilitate scientific discovery. Among these methods, deconvolution techniques are straightforward, yet either require known prior information or struggle to tackle large experimental data. Deep learning (DL)-based super-resolution (SR) methods handle large data well, however data scarcity and model generalizability are problematic. In addition, current image restoration methods are mostly offline and inefficient for many beamlines where high data volumes and data complexity issues are encountered. To overcome these limitations, an online image-restoration pipeline that adaptably selects suitable algorithms and models from a method repertoire is promising. In this study, using both deconvolution and pretrained DL-based SR models, we show that different restoration efficacies can be achieved on different types of synchrotron experimental data. We describe the necessity, feasibility, and significance of constructing such an image-restoration pipeline for future synchrotron experiments.