Soft Clustering to Determine Ambiguous Regions during Medical Images Segmentation

摘要

Image segmentation is an essential step in almost all image processing applications and very critical particularly for medical images. Image segmentation procedure segments an image into appropriate number of regions. Several techniques have been proposed and experimented to obtain effective image segmentation. Clustering is one of the commonly used image segmentation techniques. There exist ambiguous regions in an image and segmenting these regions correctly is a challenging task. Different clustering approaches are explored by researchers to deal these ambiguous regions in order to obtain better image segmentation. We propose rough clustering approach to explicitly determine ambiguous regions from an image. Once ambiguous regions are identified segmentation would be easier. In this paper we present our experiments of image segmentation using crisp K-means clustering algorithms and rough K-means (RKM) clustering algorithms. With the help of various images we demonstrate that RKM algorithm is able to determine ambiguous regions distinctly whereas K-means forced pixels of ambiguous regions to either region. Furthermore, we analyze how other soft clustering techniques deals with ambiguous regions.