Exploring the Parameter Space of Point Spread Function Determination for the Scanning Electron Microscope—Part II: Effect on Image Restoration Quality. (October 2019)

• Record Type: Journal Article
• Authors: Nevins, Mandy C.; Hailstone, Richard K.; Lifshin, Eric
• Is Part Of: Microscopy and microanalysis. Volume 25:Number 5(2019)
• Page Start: 1183
• Page End: 1194
• Publication Date: 2019-10
• Subjects: deconvolution, -- image restoration, -- low voltage, -- point spread function, -- scanning electron microscopy; Microscopy -- Periodicals; Microchemistry -- Periodicals; 502.82
• DOI: 10.1017/S1431927619014831
• Languages: English
• ISSNs: 1431-9276

Abstract

Abstract: Point spread function (PSF) deconvolution is an attractive software-based technique for resolution improvement in the scanning electron microscope (SEM) because it can restore information which has been blurred by challenging operating conditions. In Part 1, we studied a modern PSF determination method for SEM and explored how various parameters affected the method's ability to accurately estimate the PSF. In Part 2, we extend this exploration to PSF deconvolution for image restoration. The parameters include reference particle size, PSF smoothing ( K ), background correction, and restoration denoising ( λ ). Image quality was assessed by visual inspection and Fourier analysis. Overall, PSF deconvolution improved image quality. Low λ enhanced image sharpness at the cost of noise, while high λ created smoother restorations with less detail. λ should be chosen to balance feature preservation and denoising based on the application. Reference particle size within ±0.9 nm and K within a reasonable range had little effect on restoration quality. Restorations using background-corrected PSFs had superior quality compared with using no background correction, but if the correction was too high, the PSF was cut off causing blurrier restorations. Future efforts to automatically determine parameters would remove user guesswork, improve this method's consistency, and maximize interpretability of outputs.