Morphological segmentation of textured cell images

An iterative morphological algorithm for segmentation of textured cell images encountered in cytology is introduced. The initial image segmentation is determined by classifying the image local variation information obtained with dilation and erosion operations. A median filter is then used to smooth the initially segmented image. A modified median filter is applied subsequently to correct possible classification errors inside the cells. The modified median filter can clear small regions of misclassified pixels while avoiding significant changes to the cell profiles. An erosion operation is finally used to restore the cell regions. Experimental results demonstrate that the proposed algorithm is successful. Comparison of the proposed algorithm and the K-means clustering method is provided.