Robust Autofocusing for Automated Microscopy Imaging of Fluorescently Labelled Bacteria

In this paper, we present a robust focus algorithm tailored to the requirements of automated microscopy of fluorescently labelled bacteria. We combine two strategies to make the algorithm robust against focus errors due to low bacteria densities and the presence of dirt. First, we acquire a number of focus estimates for different fields on each slide well using a correlation-based focus measure. A planar model is fitted to these initial estimates using robust regression. The fit parameters provide information about slide tilt and the variability of the focus positions. We use this to constrain the search range to the locally optimal range for each field. Second, we introduce a heuristic pre-filtering step, based on a morphological area opening, into the computation of the focus measure to suppress false maxima arising from the presence of dirt particles. The effectiveness of both steps is demonstrated by focus results obtained for a sample slide.