Segmentation of cervical cancer images using active contour models

A new technique for automatic extraction of object region and boundary from the background for cell nucleus segmentation of cervical cancer images is proposed in this work. Gradient magnitude and directional information are employed to extract the exact boundary of the object under consideration. Segmentation process begins with preprocess as computation of optimum threshold based on the clusters automatically from K-means clustering algorithm. The cluster center of this threshold region, act as a seed for further processing. The problem of extracting the gradient information from concave boundaries is addressed by the use of gradient vector flow snake. The active contour has been found using convergence Index filter to find the exact boundary using directional information if the magnitude differentiation is less in the low intensity image like cell medical images. Then the object region is extracted from the object boundary and gray scale cluster. The advantage of this method is that it does not require any initial approximate contour or any seed point like parametric active contour model or any priori knowledge about the image characteristics. Further the segmentation speed of this algorithm also empowers real time execution possible for any generalize image applications.