High-resolution optical images can be attained by techniques of image reconstruction with an appropriate degradation function, which is also known as the point spread function (PSF). Currently the PSFs of a microscope is mainly obtained by measuring the images of a sample with known size. However, due to the limited depth-of-field (DOF) of optical microscopies and complicated influence of different internal and external factors during imaging, blurred and unstable optical images are easily to obtained, resulting into low precision of PSF extraction. Therefore, in this study, a blurring feature analysis system including an ideal blurring image generation module, a blurring kernel extraction module, and an image deconvolution module, based on StyleGAN was proposed based on the images of a Gaussian beam source. Then, a high-resolution image reconstruction method based on the extracted blurring kernel and the learnable convolutional half-quadratic splitting and convolutional preconditioned Richardson (LCHQS-CPCR) neural network model was proposed. Experiment with dynamic and large-scale microspheres under visible light condition demonstrated the effectiveness and correctness of our system.
Paper
Full text
Blurring Feature Analysis of Microscopic Images Based on Deep Learning
Semantic Scholar · Computer Science · 2021
Abstract
High-resolution optical images can be attained by techniques of image reconstruction with an appropriate degradation function, which is also known as the point spread function (PSF). Currently the PSFs of a microscope is mainly obtained by measuring the images of a sample with known size. However, due to the limited depth-of-field (DOF) of optical microscopies and complicated influence of different internal and external factors during imaging, blurred and unstable optical images are easily to obtained, resulting into low precision of PSF extraction. Therefore, in this study, a blurring feature analysis system including an ideal blurring image generation module, a blurring kernel extraction module, and an image deconvolution module, based on StyleGAN was proposed based on the images of a Gaussian beam source. Then, a high-resolution image reconstruction method based on the extracted blurring kernel and the learnable convolutional half-quadratic splitting and convolutional preconditioned Richardson (LCHQS-CPCR) neural network model was proposed. Experiment with dynamic and large-scale microspheres under visible light condition demonstrated the effectiveness and correctness of our system.