Coarse-to-Fine Particle Segmentation in Microscopic Urinary Images

This paper presents a coarse-to-fine particle segmentation strategy to extract particles from microscopic urinary images within two stages, coarse stage and fine stage. In coarse stage, to locate particles in a wide range of images including the low contrast, the unevenly illuminated, etc, we develop 4-direction variance mapping followed by an adaptive thresholding method. Within this stage, particles are well located, but their contours fail to exactly represent their shapes and clumped particle clusters are not divided. In fine stage, combined with Canny edges, we extract desired particle contours, then an effective local maxima search algorithm based on distance map successfully separates clumped particle clusters into individual particles. Our strategy is easy for implementation and its effectiveness is verified by large-scale experiments.