一种基于改进径向基神经网络的人脸图像识别方法

常用于径向基神经网络中心参数学习的K-均值聚类算法,易受初始参数选取的影响而收敛于局部极小值.将自动终止聚类判据的减聚类算法用于径向基网络的学习,可根据样本集确定径向基函数数目,且其计算量与数据点的数目与考虑问题的维数无关,很适合于人脸这种维数较高的模式.实验证明,应用这种算法训练径向基神经网络识别人脸,从识别精度到识别速度上都优于传统算法.

Face Recognition Method Based on Improved Radial Basis Function Neural Networks

Influenced by initial selected value,K-means clustering algorithm applied to training the center parameters of radial basis function(RBF) neural networks tends to converge to a local minimum.The subtractive clustering algorithm can automatically terminate the clustering criterion to ascertain RBF numbers.By using the algorithm to train RBF networks,the number of data points and their calculation are irrelevant to the dimension of the information considered.For a high-dimensional face pattern,the algorithm can effectively pick up training speed.The results show that RBF neural networks classifier with the proposed algorithm is more effective in speed and accuracy than that with the traditional algorithm.