Non-iterative Blind Deblurring of Digital Microscope Images with Spatially Varying Blur

Furkan Kaynar, Peter Geißler, Laurent Demaret, Tamara Seybold, Walter Stechele

Abstract: One of the main limiting factors of image quality in surgical microscopy is of physical nature: resolution is limited by diffraction effects. The digitalisation of surgical microscopy allows computational solutions to partially compensate for this limitation of the involved optics. An inherent characteristic of microscope optics is that it is diffraction-limited which leads to blurred images of objects that do not lie in the (often very narrow) focus plane. Digital deblurring techniques can correct this during the surgical operation, however the point spread function is not constant spatially, making the problem complicated and extremely ill-posed. Most blind deblurring algorithms formulate an iterative solution to estimate the latent sharp image, which is not appropriate for processing high-resolution, high frame rate videos in real-time conditions. We propose a novel single-pass non-iterative blind deblurring method which estimates the spatially varying point spread function by evaluating structural details locally and performing deblurring only at pixels with significant structural information, avoiding noise amplification and decreasing computational cost. The quantitative and qualitative experiments showed the effectiveness and robustness of our method, indicating the promising nature of image enhancement for microscopy-based surgical operations.