Abstract-A new image segmentation using multi-hyperbolic and multi-Gaussian kernel based fuzzy c-means algorithm (MHMGFCM) is proposed for medical magnetic resonance image (MRI) segmentation. The integration of two hyperbolic tangent kernels and two Gaussian kernels are used in the proposed algorithm for clustering of images. The performance of the proposed algorithm is tested on OASIS-MRI image dataset. The performance is tested in terms of score, number of iterations (NI) and execution time (TM) under different Gaussian noises on OASIS-MRI dataset. The results after investigation, the proposed method shows a significant improvement as compared to other existing methods in terms of score, NI and TM under different Gaussian noises on OASIS-MRI dataset.Keywords-FCM, multi-hyperbolic tangent function, Segmentation, multi-Gaussian Kernal, fuzzy, multiple-kernal.
I. INTRODUCTIONIn image processing and computer vision, medical image segmentation is an active research area [1]. The process of clustering the image into non-overlaped, consistent regions is called the image segmentation. These regions are identical with respect to some features like texture, color, shape, intensity etc. Based on the features, the process of segmentation is separated into four groups: clustering (intensity), thresholding (intensity), region extraction (color or texture) and edge detection (texture).In Clustering is a procedure for classifying patterns or objects in such a manner that samples of the same cluster are more comparable to one another than samples belonging to other clusters. There are two main clustering approachs: the hard clustering technique and the fuzzy clustering technique. MacQueen [8] has proposed the k-means clustering algorithm. The k-means is one of the hard clustering technique. The usual hard clustering techniues classify every point of the records set just to one cluster. As a effect, the results are often very crusty, i.e., in image clustering every pixel of the image goes to one cluster. However, in many real conditions, issues such as restricted spatial resolution, reduced contrast, partly cover intensities,