An improved xie-beni index for cluster validity measure

The pathology may appear as a new cluster(s) on radiological images and hence the information of cluster location cannot decide in prior. In this regard, the unsupervised methods of segmentation play the important role, however, these methods need the number of clusters as the input. The challenging tasks in clustering based image segmentation are to choose the number of segments in an image. This work proposes the segmentation quality index, which utilizes the trend of Xie-Beni index to obtain the optimum number of segments in an image. The proposed algorithm has been implemented on the segmentation results obtained by enhanced fuzzy c-means algorithm and compared with the classical validity indexes such as Xie-Beni index, partition entropy coefficient, partition coefficient and fuzzy hyper-volume on synthetic images and simulated brain MRI dataset images. The quantitative results show that the proposed method has greater ability to find the appropriate number of clusters on the ground truth and noisy images.