Yeast cell detection and segmentation in bright field microscopy

We present a method for detecting and segmenting yeast cells in bright field microscopy images from which cells are often almost transparent. A classifier is firstly trained to detect edges of cells of interest. A label cost model with cardinality constraints then simultaneously detects cell centers and clusters cell edge points, using Integer Linear Programming. For a noisy or partial edge clustering, an additional step of contour fitting or seeded watershed is applied for segmentation. Results demonstrate that our method can consistently detect and segment yeast cells from a variety of datasets, and its performance is close to that of manual segmentation.