基于改进LBP和DBN的人脸识别算法研究  

Study on face recognition algorithm based on improved LBP and DBN

作　　者：吕秀丽[1] 黄兆昊 白永强[1] LÜ Xiuli;HUANG Zhaohao;BAI Yongqiang(Northeast Petroleum University,School of Physics and Electronic Engineering,Heilongjiang Daqing 163318,China)

机构地区：[1]东北石油大学物理与电子工程学院,黑龙江大庆163318

出　　处：《工业仪表与自动化装置》2021年第5期80-82,共3页Industrial Instrumentation & Automation

基　　金：黑龙江省自然科学基金资助(LH2019D006)。

摘　　要：针对传统LBP算法的人脸识别易受光照、背景、遮挡等因素的影响,使用改进局部二值模式(LBP)和深度信念网络(DBN)相结合的方法,用多尺度块局部二值模式(MB-LBP)算法获取人脸图像的纹理特征,在此人脸纹理特征的基础上使用中心对称局部二值模式(CS-LBP)算法获取图像的纹理特征,然后将两次获得的纹理特征图像的直方图进行融合,并将其输入到DBN中进行训练,优化网络参数。The traditional LBP algorithm for face recognition is easily affected by illumination,background,occlusion and other factors.Using the improved local binary pattern(LBP)and depth belief network(DBN)method,the texture features of face image are obtained by using multi-scale block local binary pattern(MB-LBP)algorithm.On the basis of the face texture features,the centrosymmetric local binary pattern(CS-LBP)is used the algorithm obtains the texture features of the image,and then fuses the histograms of the two texture feature images,and inputs them into the DBN for training to optimize the network parameters.

关 键 词：局部二值模式 纹理特征 特征融合 深度信念网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]