Novel region labeling method for cellular images

Cellular images are the challenges in the field of image segmentation because of their intrinsic properties. A dual-layer structure was adopted in order to solve this problem: the 1st layer based on traditional Split-Merge method introduces and simplifies pulse coupled neural network to split the images, and Mumford-Shah model to merge the split areas. The output is a coarse segmented image. The 2nd layer extracts the edges via Canny operator and regards the edges whose length are larger than given threshold as cell edge while the opposite as false-edges caused by noises. The output is a non-continuous edge. The two layers are combined together to obtain the final labeling results via skeleton extraction and thinning from mathematical morphology. Experiments with 5 different cell images demonstrate the proposed algorithm outperforms than region-based segment and a priori model-based segment on the basis of the discussion on parameters selection and comparison with those methods in terms of correct labeling rate and computation time.