Abstract: Overlapping cell images with inhomogeneity intensity, low contrast and edge blurring are difficult to be segmented. The author proposes a new cell segmentation algorithm combining level set method and concave area detection. First, the level set method can easily handle topology changes of the evolving contour. And it can be employed to obtain the cell profile, combined with the regional information and edge information. This step keeps the geometric characteristics of the cell profile effectively. Second, the concave area of overlapping contour location based on the concave and convex of polygons was searched for. Finally, the splitting line of overlapping cells at the location of concave area was determined. Experiments on dozens of different overlapping cell images segmentation show that the algorithm is robust, effective and easy to implement. The average accuracy of cell segmentation reaches to 83.01%, which is superior to the results of the watershed and k-means clustering methods.

Key words: cell segmentation, level set method, overlapping cells, polygonal concave and convex