三种面部性别识别算法的比较  

Comparison of three facial gender recognition algorithms

作　　者：李宇 丁文倩 金立左[1] 潘泓[1] Li Yu;Ding Wenqian;Jin Lizuo;Pan Hong(School of Automation,Southeast University,Jiangsu Nanjing,210096,China)

机构地区：[1]东南大学自动化学院,江苏南京210096

出　　处：《机械设计与制造工程》2018年第12期37-42,共6页Machine Design and Manufacturing Engineering

基　　金：国家自然科学基金面上项目(61671151)

摘　　要：首先介绍了FASTA-ELM、多尺度融合决策和多量化局部二进制模式三种算法在面部性别识别领域的应用。然后选取了FERET和PAL两个人脸数据集,其中PAL中的人脸图像由于角度和光照的影响,有很大的识别难度。接着将三种算法在这两个数据集上分别进行测试,发现FASTA-ELM能在不使用SVD的情况下计算输出权重的最小范数,并且可以推广到隐藏层矩阵较大的情况中去;而多尺度融合决策由于融合了多个尺度的信息,可以改善识别效果;多量化局部二进制模式则具有更好的泛化能力和处理噪声的能力。三种算法都达到了90%以上的识别正确率。It introduces the application of FASTA-ELM,multi-scale fusion decision-making and multi-quantization local binary mode in facial gender recognition.Taking two face data sets of FERET and PAL,it shows that the face image in PAL has great recognition difficulty due to the influence of angle and illumination.Then the three algorithms are tested on the two data sets respectively.It is found that FASTA-ELM can calculate the minimum norm of the output weight without using SVD,and can be extended to the case where the hidden layer matrix is large;multi-scale fusion decision-making can improve the recognition effect by combining multiple scales of information;multi-quantization local binary mode has better generalization ability and ability to process noise.All three algorithms can achieve a correct rate of more than90%.

关 键 词：面部性别识别 FASTA-ELM 多尺度融合决策 多量化局部二进制模式 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]