Abstrato

A review on optimized K-means and FCM clustering techniques for biomedical image segmentation using level set formulation

Chenigaram Kalyani, Kama Ramudu, Ganta Raghotham Reddy

Image segmentation plays a vital role in a research area of computer vision as well as image processing. One of the most popularly used image segmentation techniques are named as Clustering. It is a process of segmenting a non-homogenous image intensities into a homogeneous regions based on their similarity function. But, here the clustering techniques need a prior initialization of random cluster centers, which is a primary disadvantage. So, in order to get an “optimum” cluster centers we need to initialize a PSO algorithm in pre-processing an image. In this way successfully we can stay away from the drawback of sensitivity of initial esteems. Algorithm of PSO clustering, were widely utilized in recognition of pattern approaches just like an image segmentation, after initializing the PSO algorithm, it improves the clustering efficiency but we do have increasing of computational time based on the population size. At the same time few boundary leakages i.e., outliers presented in clustered outputs. So, in post-processing, we were introducing a level set formulation (LSF) to overcome the boundary leakages which were occurred in pre-processing of an image. Here in this paper, we give a brief description on few clustering methods at the same time some of the recent works by researchers on these methods and also we present K-means in addition with Fuzzy C-means (FCM) clustering algorithm along with the Particle Swarm Optimization an application to a biomedical image for an efficient segmentation of the biomedical images. The performances of the segmentation outputs are compared using Dice similarity and Jaccard similarity values and it validate the effectiveness of the proposed method.