【作者】 李翠华； 郑南宁； 张永平；

【Author】 Li Cuihua Zheng Nanning Zhang Yoagping (Institute of Artificial Intelligence and Robotics,Xi’an Jiaotong University,Xi’an 710049)

【机构】 西安交通大学人工智能与机器人研究所；

【摘要】 本文以二维Gauss函数的卷积逼近公式为基础,构造了一类中心固定的径向基函数网络, 并以最小二乘(LS)为准则,给出了基于低松弛迭代格式的快速学习算法．根据Gauss函数的速降性,可以对网络采用稀疏连接,对于一个N×N输入图象。学习算法的计算复杂度为O(N2), 就LS算法的复杂性而言,比文中不动点(FP)迭代算法的O(N3)及文中基于FFT的预处理共轭梯度(PCG)算法的O(N2·logN)都好．该网络可以在计算机视觉及高维信号处理等领域用作自适应滤波．实验结果表明。它对Gauss白噪声及均匀分布的噪声都具有良好的抑制特性．更多还原

【Abstract】 Based on an expansion of Gaussian function and the law of least squares, the radial basis function network with underrelaxation iterative leaming algorithm for the removal of noise from image is proposed In this paper. Fora NxN image, the computational complexity of the learning algorithm is O(N2), which is significantly better than O(N3) using fixed point iterative method for handling LS problem in paper and O(N2 log N) using the preconditioned conjugate gradient iterative algorithm in paper . Consequently, the radial basis function network may be used as an adaptive filter in computer vision and realtime signal processing. The test results show that the radial basis function network is suited for either Gaussian noise reduction or the noise with uniform distribution.更多还原