An Efficient K-Means Clustering For Image Segmentation With An Application To MRI

Segmentation means to classify the objects that exist in an image. One natural view of segmentation is an attempt to determine which components of data set naturally belong together. This is a problem known as clustering. Clustering techniques falls into several categories such as neural network based, center based, region growing etc. Among the existing techniques, Center based techniques is commonly used for segmentation. K. Means Clustering and its variations such as Fuzzy K Means, K- Harmonic Means, and Relational K Means etc, are the examples of center based techniques. The variations are based on proper choice of cluster memberships, initial values for cluster center and the weights assigned to pixels for participating in the cluster. This paper proposes an efficient variant of k-means clustering algorithm which could produce quality segments with predictable performance. The proposed algorithm could cluster the input image in three scans. The algorithm is applied on color images for segmentation, also on MRI images to estimate the cortical gray matter volume on a semi- automatic basis. The results obtained are found to be better while comparing with results produced by the other variants of k-means, namely Hybrid 1 and Hybrid 2.