基于LBP和深度学习的非限制条件下人脸识别算法

doi:10.3969/j.issn.1000-436x.2014.06.020

摘要/Abstract

摘要：

提出一种在非限制条件下，基于深度学习的人脸识别算法。同时，将LBP纹理特征作为深度网络的输入，通过逐层贪婪训练网络，获得良好的网络参数，并用训练好的网络对测试样本进行预测。在非限制条件下人脸库LFW上实验结果表明，该算法较传统算法（PCA、SVM、LBP）识别率高；另外，在Yale库和Yale-B库上也获得较高识别率，进一步说明以LBP纹理特征作为网络输入的深度学习方法能够对人脸图像进行准确识别。

关键词: 非限制条件, 人脸识别, LBP, 深度网络, 深度学习

Abstract:

A face recognition method under unconstrained condition was proposed based on deep learning. At the same time, making LBP texture features as the input of deep learning net, and greedy training the network layer was made by layer to obtain good network parameters. At last, the trained net was used to predict the test samples' labels. The results of experiments on LFW(labeled faces in the wild) show that the algorithm can obtain higher recognition rate than traditional algorithms(such as PCA, SVM, LBP).Otherwise, the recognition rate on Yale and Yale-B are also very high, the experi-mental results show that deep learning net with LBP texture as its input can classify face images correctly.

Key words: unconstrained condition, face recognition, LBP, deep network, deep learning

梁淑芬,刘银华,李立琛. 基于LBP和深度学习的非限制条件下人脸识别算法[J]. 通信学报, 2014, 35(6): 154-160.

Shu-fen LIANG,Yin-hua LIU,Li-chen LI. Face recognition under unconstrained based on LBP and deep learning[J]. Journal on Communications, 2014, 35(6): 154-160.

图/表 9

图1

图2

表1

表2

表3

表4

表5

表6

表7

参考文献 17

[1]	MEDIONI G , CHOI J , KUO C H , et al. Identifying noncooperative subjects at a distance using face images and inferred three dimensional face models[J]. IEEE Trans Syst, Man, Cybern A, Syst, Humans, 2009,39(1): 12-24.
[2]	BLANZ V , VETTER T . Face recognition based on fitting a 3D morphable model[J]. IEEE Transaction on Pattern Analysis and Ma-chine Intelligence, 2003,25(9): 1063-1074.
[3]	LIOR W , TAL H , YANIV T . Effective uncon-strained face recognition by combining multiple descriptors and learned background statistics[J]. IEEE Pattern Analysis and Machine Intelligence, 2011,33(10): 1978-1990.
[4]	MARSICO M D E , NAPPI M , RICCO D . Robust face recognition for uncontrolled pose and illumination changes[J]. IEEE Transactions on Systems, Man and Cybernetic, 2012,43(1): 149-163.
[5]	JAVIER R , RODRIGO V , MAURICIO C . Recognition of faces in unconstrained environments: a comparative study[J]. Journal on Ad-vances in Signal Processing. 2009,12(4): 44-69.
[6]	WOLF L , HASSNER T , TAIGMAN Y . Descriptor based methods in the wild[A]. Faces in Real-life Images Workshop in ECCV[C]. 2008.1-14.
[7]	ZHAO D , LIN Z , XIAO R , et al. Linear laplacian discrimination for feature extraction[A]. Proc IEEE Conference on Computer Vision and Pattern Recognition[C]. 2009.1-7.
[8]	BENGIO Y , DELALLEAU O . On the expressive power of deep archi-tectures[A]. Proc of 14th International Conference on Discovery Sci-ence[C]. Berlin: Springer-Verlag, 2011.18-36.
[9]	HINTON G E , OSINDERO S , THE Y-W . A fast learning algorithm for deep belief nets[A]. Neural Computation, 2006,18(7): 1527-1554.
[10]	COTTRELL G W . New life for neural networks[J]. Science, 2006,313(5786): 454-455.
[11]	TALOR G W , HINTON G E , ROWEIS S T . Modeling human motion using binary latent variables[A]. Advances in Neural Information Processing Systems[C]. 2007.1345-1352.
[12]	ITAMAR A , DEREK C R , THOMAS P K . Deep machine learning—a new frontier[J]. Artificial Intelligence Research IEEE Computa-tional Intelligence Magazine, 2010,5(4): 13-18.
[13]	OJALA T , PIETIKAINEN M , MAENPAA T . Multiresolution gray-scale and rotation invariant texture classification with local binary patterns[J]. IEEE Transactions on Pattern Analysis and Machine Intel-ligence, 2002,24(7): 971-987.
[14]	ZHAO Z , PIETIKAINEN M . Dynamic texture recognition using local binary patterns with an application to facial expressions[J]. IEEE Trans PAMI, 2007,27(6): 915-928.
[15]	LEI Z , LIAO S . Face recognition by exploring information jointly in space, scale and orientation[J]. IEEE Trans on Image Processing, 2009,2(1): 1-127.
[16]	LEI Z , LIAO S . Face recognition by exploring information jointly in space, scale and orientation[J]. IEEE Trans on Image Processing, 2009,2(1): 1-127.
[17]	HINTON G E . Face recognition by exploring information jointly in space, scale and orientation[J]. Neural Computation, 2002,14(8): 1771-1800.

方法	隐藏单元数	识别率	训练时间/s	分类时/s	总时间/s
	1 000	24.83%	149.17	6.36	155.53
	2 000	45.15%	383.52	15.81	399.33
DBN	3 000	61.45%	720.31	30.34	750.65
	4 000	74.57%	1 164.53	45.39	1 209.92
	5 000	81.30%	1 706.08	67.48	1 773.56
	1 000	24.24%	471.84	13.05	484.89
传统LBP+DBN	2 000	45.56%	859.80	22.86	882.66
	3 000	61.65%	2 745.55	78.59	2 824.14
	1 000	25.88%	156.58	6.72	163.3
	2 000	47.92%	403.27	17.30	420.57
LBP+DBN	3 000	62.63%	766.61	28.45	795.06
	4 000	75.62%	1 173.92	47.30	1 221.22
	5 000	82.77%	1 743.00	68.06	1 811.06

方法		训练样本数
方法
	5	10	15
PCA	31.92%	38.98%	39.46%
SVM	36.58%	48.29%	52.40%
LBP	38.64%	51.43%	56.27%
DBN	81.30%	82.59%	82.99%
LBP+DBN	82.77%	84.02%	84.54%

方法	识别率
PCA	35.93%
SVM	45.51%
LBP	47.30%
FACE	61.00%
DBN	84.05%
LBP+DBN	85.16%

方法	1 000	2 000	3 000	4 000	5 000
DBN	44.21%	67.78%	80.48%	90.00%	96.67%
LBP+DBN	46.84%	68.29%	81.89%	90.14%	98.89%

方法	识别率
PCA	80.00%
SVM	94.44%
LBP	93.00%
DBN	96.67%
LBP+DBN	98.89%