Watershed Segmentation of Cervical Images Using Multiscale Morphological Gradient and HSI Color Space

In this paper, a novel watershed segmentation algorithm is proposed for the segmentation of nucleus from the surrounding cytoplasm of cervical cancer images. The proposed method converts the input RGB image into HSI image which contains three components hue, saturation and intensity. The saturation component is thresholded to obtain the binary image and each pixel in the binary image is multiplied with hue component to obtain the product image. The intensity image is complemented, thresholded and merged with the product image and smoothened. The local minima are reduced using extended minima function and the multiscale gradient of this resultant gray scale image is segmented using watershed algorithm based on Hill Climbing technique. Experimental results add to the computational efficiency of the algorithm, its shape maintaining, edge preserving and scale-calibrating features. The performance is also superior to most other segmentation techniques.