Frequency domain kurtosis-based no-reference image quality assessment for bright-field microscopy images

In the last few years, image processing researchers spent a substantial amount of time and effort developing and perfecting image quality assessment algorithms. Bright-field microscopy, for example, produces images whose quality is a bottleneck for consistent evaluation. For instance, when a stack of images of a specimen is acquired in different focal plane configurations, there will be a set of blurred or partially blurred elements in it, impairing proper evaluation. This work aims to provide an image quality assessment metric, without the presence of a reference image for comparison, to detect the blurred and sharp images among the whole set of the stack, and elect the sharpest ones for a further fusion process. The correlation of the results with subjective labeling of the image sets showed that the proposed metric offers reliable identification of the eligible images for fusion and suggests the application in other real-world problems.