Texture image classification using modular radial basis function neural networks

摘要

Image classification has become an important topic innmultimedia processing. Recently, neural network-based methodsnhave been proposed to solve the classification problem. Amongnthem, the radial basis function neural network (RBFNN) is the mostnpopular architecture, because it has good learning and approxima-ntion capabilities. However, traditional RBFNNs are sensitive to cen-nter initialization. To obtain appropriate centers, it needs to find sig-nnificant features for further RBF clustering. In addition, the trainingnprocedure of a traditional RBFNN is time consuming. Therefore, innthis work, a combination of a self-organizing map (SOM) and learn-ning vector quantization (LVQ) neural networks is proposed to selectnmore appropriate centers for an RBFNN, and a modular RBF neuralnnetwork (MRBFNN) is proposed to improve the classification ratenand to speed up the training time. Experimental results show thatnthe proposed MRBFNN has better performance than those of thentraditional RBFNN, the discrete wavelength transform (DWT)-basednmethod, the tree structured wavelet (TWS), the discrete waveletnframe (DWF), the rotated wavelet filter (RWF), and the wavelet neu-nral network based on adaptive norm entropy (WNN-ANE)nmethods.