Kernel-Based Centroid Neural Network with Spatial Constraints for Image Segmentation

Abstract

A kernel-based centroid neural network with spatial constraints (K-CNN-S) is proposed and presented in this paper. The proposed K-CNN-S is based on the centroid neural network (CNN) and also exploits advantages of the kernel method for mapping input data into a higher dimensional feature space. Furthermore, The K-CNN-S adopts the spatial constraints to reduce noise in images. The magnetic resonance image (MRI) segmentation is performed to illustrate the application of the proposed K-CNN-S algorithm. Experiments and results on MRI data from Internet brain segmentation repository (IBSR) demonstrate that image segmentation scheme based on the proposed K-CNN-S outperforms conventional algorithms including fuzzy c-means (FCM), kernel-based fuzzy c-mean (K-FCM), and kernel-based fuzzy c-mean with spatial constraints (K-FCM-S).