Super-Resolution Reconstruction Based on Denoised High-Resolution Raw Images with BM3D

Abstract

The pixel in a conventional raw image (CR) and the point spread function’s standard deviation of the microscope are approximately equal in size. A high-resolution raw image (HR) lacks research value due to excessive noise. Its pixel size is only half that of CR. BM3D is an excellent denoising algorithm. We propose a super-resolution microscopy method. It denoises HR and uses compressed sensing for super-resolution reconstruction. It was compared with that of HR before denoising, and CR before and after denoising. HR and CR with three different noise levels (low, medium, and high) are studied in simulation. Simulation results demonstrate that BM3D is not only related to the noise type and the noise level, but also to the raw image’s pixel size. In the medium noise level, denoised HR performed the best super-resolution reconstruction, followed by denoised CR. Real experiment results are closer to the simulation results in the medium noise level.