Segmentation of Overlapping Cervical Cells in Microscopic Images with Superpixel Partitioning and Cell-Wise Contour Refinement

Hansang Lee, Junmo Kim

Abstract: Segmentation of cervical cells in microscopic images is an important task for computer-aided diagnosis of cervical cancer. However, their segmentation is challenging due to inhomogeneous cell cytoplasm and the overlap between the cells. In this paper, we propose an automatic segmentation method for multiple overlapping cervical cells in microscopic images using superpixel partitioning and cell-wise contour refinement. First, the cell masses are detected by superpixel generation and triangle thresholding. Then, nuclei of cells are extracted by local thresholding and outlier removal. Finally, cell cytoplasm is initially segmented by superpixel partitioning and refined by cell-wise contour refinement with graph cuts. In experiments, our method showed competitive performances in two public challenge data sets compared to the state-of-the-art methods.