基于神经网络的快速核子空间人脸识别算法研究

针对现有核子空间人脸识别算法计算量大且速度缓慢的现状,提出了一种基于神经网络的快速核子空间人脸识别算法模型,利用神经网络的隐含层神经元将核特征子空间的基表示进行约减,从而大幅提高了识别速度。进而基于KPCA和KFDA两种核子空间人脸识别算法,建立了神经网络逼近模型,并基于ORL、UMIST和YALE 3种人脸数据库进行了实证分析。实验结果表明,当隐含层神经元个数设置为训练样本总数一半或更少时,基于神经网络的快速核子空间算法能够取得相近甚至相当于核子空间算法的识别率。从而在满足一定识别正确率的条件下,能将识别时间缩短到50%甚至更低。

Fast Kernel Subspace Face Recognition Algorithm Based on Neural Network

Aiming at the situation of large calculation amount and slow calculation speed of existing kernel subspace face recognition algorithm,this paper presented a fast kernel subspace face recognition algorithm based on neural network model,and used neuron of neural network hidden layer to reduce base representation of nuclear feature subspace in order to improve the speed of recognition.Then we established the neural network approximation model based on the KPCA and KFDA kernel subspace face recognition algorithms and made an analysis on the database of ORL,UMIST and YALE.The experimental results show that when the number of the hidden layer neuron is set as half of the training samples or less,the fast kernel subspace algorithm based on neural network can achieve similar or even equal recognition rate to the nuclear subspace algorithm,and the recognition time can be reduced to 50% or even lower with certain correct recognition rate.