The good, the bad and the ugly : droplet recognition by a 'shootout'-heuristics

We explain how to optimize the image analysis of mixed clusters of red and green droplets in solvents with various degrees of sharpness, brightness, contrast and density. The circular Hough Transform is highly efficient for separated circles with reasonable background contrast, but not for large amounts of partially overlapping shapes, some of them blurred, as in the images of our dense droplet suspensions. We explain why standard approaches for image improvement fail and present a “shootout” approach, where already detected circles are masked, so that the removal of sharp outlines improves the relative optical quality of the remaining droplets. Nevertheless, for intrinsic reasons, there are limits to the accuracy of data which can be obtained on very dense clusters.