Blurring Degree Calculation from Single Microscopic Defocused Image Based on Adaptive Spectral Clustering

Deconvoluting a defocused image with a degradation function, also named as point spread function, is an effective method for image de-blurring. However, the blurring kernel of a point spread function is highly susceptible to various disturbances such as image texture, brightness, and noise due to the complex imaging mechanism in high-magnification optical microscopes, and it is difficult to fast and accurately obtain the point spread function of a defocused microscopic image. In this paper, the relationship between point spread function in the blurring imaging process and image similarity function in adaptive spectral clustering (ASC) is analyzed, and a defocused image classification method according to the blurred degree of each image is proposed based on ASC. Subsequently, a normalized blurring degree estimation method for single microscopic blurred image is proposed based on ASC combined with the K-nearest neighbor method, and the point spread function denoted as a Gaussian function can be obtained with the blurred kernel value. Finally, experiments based on the nanoscale square grid proves the effectiveness and robustness of the proposed method.