材料期刊网

为了提高人脸识别正确率,提出基于改进非负矩阵分解的神经网络人脸识别算法.首先利用改进的非负矩阵分解对人脸图像进行特征提取,提高非负矩阵分解速度.接着将提取出的特征信息作为神经网络学习入口进行特征训练,由于神经网络在学习过程中,容易出现局部最小值且收敛速度慢等问题,为此采用改进的遗传算法对神经网络进行优化处理,获得最终的人脸识别结果.实验结果表明:利用改进的非负矩阵分解方法能够降低神经网络的分类训练负荷量和运算量,提高人脸识别识别率.通过和各种方法比较可知,本方法的人脸识别率都较高.本方法人脸特征分解速度快,提高了神经网络训练前期精度和收敛速度,使得人脸识别正确率高.当特征向量个数达到40以上时,人脸识别正确率保持95%以上.

In order to promote the accuracy of facial recognition, an improved algorithm of the neural network based on NMF for recognizing faces was proposed.First, features from facial images are extracted by using the improved method of non-negative matrix factorization to increase the decomposition speed of NMF.Then the extracted feature information is made as a neural network learning entrance for characteristics training.Due to the problems of minimum local value and slow convergence speed in the process of neural network learning, an improved genetic algorithm was used to optimize the neural network, and final facial recognition result was obtained.Experimental results indicate that the facial recognition method using the improved NMF can reduce the classification training load and operand of the neural network, and can also increase the facial recognition rate.Compared with other methods, our method has a higher facial recognition rate.The proposed method has high facial feature decomposition speed, it also promote the accuracy of neural network training and improve the convergence speed to make the rate of facial recognition higher.With more than 40 number of feature vectors, the facial recognition accuracy can basically keep more than 95%.