A new neural network model for medical color image segmentation
Chen Tang, Tengliang Luo, Guimin Zhang, Fang Zhang · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2006
This paper describes a new neural network model that performs color image segmentation in an unsupervised manner. The new scheme is called enhancing learning algorithm on the radial basis function neural network (ERBF). First, ERBF employs a dynamic nearest neighbor-clustering algorithm to set its front two layers: the input layer and the hidden layer. Second, ERBF network introduces the Hebb rule to train the hidden layer and divide the hidden neurons center vectors into two meaningful groups: one group members are the target color-clustering centers; the other group members are the background color-clustering centers. Finally, the ERBF output layer is trained by the competitive algorithm and outputs different values with different input values so as to divide the target region from the image. The present model avoids the trouble in deciding the hidden nodes number beforehand and needs only to be trained twice. This new color image segmentation scheme has been implemented and tested on medical color images. The results shown that the new segmentation scheme is proved to be an effective segmentation algorithm.