实时系统下LBP与CNN结合的人脸识别方法  

Face recognition based on combination of LBP and CNN in real-time system

作　　者：徐杰[1] 孙超 郭春赫 Xu Jie;Sun Chao;Guo Chunhe(School of Electronic&Information Engineering,Heilongjiang University of Science&Technology,Harbin 150022,China)

机构地区：[1]黑龙江科技大学电子信息与工程学院,哈尔滨150022

出　　处：《黑龙江科技大学学报》2018年第6期692-696,共5页Journal of Heilongjiang University of Science And Technology

基　　金：黑龙江省自然科学基金面上项目(F200921)

摘　　要：为提高实时系统下人脸识别的效果,通过具有良好尺度不变性的圆形局部二值模式直方图提取其特征,考虑计算量和光照干扰因素的影响,采用对光照有较强鲁棒性的ULBP方式对提取到的特征降维,使用直方图交叉核方式对降维后的数据计算相似度,结合被交叉熵代价函数和Adam优化器提升训练速度的卷积神经网络进行人脸识别。结果表明,通过尺度变换与有无遮挡及表情测试,在实时系统下局部二值模式直方图和卷积神经网络结合实现的人脸识别具有良好的实时性和有效性。该研究对提高人脸识别的准确性提供了参考。This paper aims to face recognition of faces detected in real-time systems.The feature extraction is performed by a circular local binary mode histogram with good scale invariance;the extracted features are considered to be computationally intensive and the effects of illumination interference factors are robust to the ULBP method for reducing the illumination;using a histogram the graph cross-core method calculates the similarity of the reduced-dimensional data and combines the convolutional neural network with the cross-entropy cost function and the Adam optimizer to improve the training speed for face recognition.Through the scale transformation and the presence of occlusion and expression tests,the face recognition realized by the combination of local binary mode histogram and convolutional neural network in real-time system has good real-time and effectiveness.

关 键 词：人脸识别 局部二值模式 卷积神经网络 交叉熵 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]