BDIC: boosting the performance of optical microscopy using blind deconvolution and illumination correction

Many computational imaging methods have been studied to increase the imaging quality for microscopy however, these
methods require special modification of the image platform and cannot be applied directly to an arbitrary microscopy
system. Inspired by the fact that the incoherent image formation is described by the convolution between the system’s
point spread function and the product between illumination pattern and sample’s intensity, we combined the single image
blind deconvolution (BD) with illumination correct (IC), and propose an image processing-based method termed BDIC to
boost the performance of an arbitrary microscopy system in a plug-and-play manner. Only the single raw image captured
by a microscope is required as input for BDIC, and it returns a blurriness-suppressed, illumination corrected enhanced
image without any modification or prior knowledge of the imaging system. In BDIC, we use the patch-wise minimum
pixel prior to achieve the blind deconvolution and we also propose a hybrid method being the combination of the Retinex
and Dark channel prior for illumination correction. We demonstrate BDIC on different samples including USAF targets,
plant root tissue and human blood smears. To evaluate its performance, we collected over 100 images for a USAF target
under different microscopy environments, capturing the photos using objective lenses with different qualities or under
good and problematic illumination conditions. Images under good microscopy conditions were set as a reference for
objective evaluation using peak signal-to-noise ratio and structural similarity. Both subjective and objective assessments show that BDIC increases the image's quality including its contrast and signal-to-noise ratio.