基于交叉覆盖神经网络的人脸识别

提出一种利用交叉覆盖神经网络(ACNN)进行人脸识别的方法.该方法应用基于几何意义的MP神经网络模型——球面邻域模型,将神经网络的训练问题转化为点集覆盖的问题.通过分析交叉覆盖算法,阐述了人脸识别的训练算法,该算法克服了传统BP算法识别率低和易陷入局部最小的问题.实验结果表明,在采用多样本训练后,该算法具有较高的识别率。

The Face Recognition Based on Alternate Covering Neural Networks

A face recognition algorithm based on alternate covering neural networks is proposed.The training problem of neural networks is transformed into the geometrical covering problem of a point set by the sphere neighborhood model,which is based on a geometrical representation of MP neural model.The training approach for face recognition is described by analyzing the alternate covering algorithm.Converging slowly and falling into the local minimum point easily in the conventional BP algorithm is avoided.Experiments show that higher recognition rate can be obtained.