基于深度学习的有遮挡人脸识别方法研究

doi:10.16180/j.cnki.issn1007-7820.2022.01.006

摘要/Abstract

摘要：

针对传统CNN在有遮挡人脸识别中计算量大的问题,文中以L1-2DPCA为基础,提出了一种用于人脸识别的新型PCANet深度学习网络。该网络以L1-2DPCA学习多个卷积层的滤波器,在卷积层之后,通过二进制散列和逐块直方图进行池化。文中以CNN、PCANet、2DPCANet和L1-PCANet作为比较,在AR和RMFD人脸数据集上测试了文中所提出的网络。结果表明,在所有测试中,文中提出的深度学习网络的识别性能优于其他网络。由于采用L1范数,文中所提出的深度学习网络对异常值和训练图像的变化具有较强的鲁棒性。

关键词: 人脸识别, 遮挡, 深度学习, L1-2DPCA, 二维主成分分析, L1范数, 卷积神经网络, 鲁棒性

Abstract:

In view of the large amount of calculation in traditional CNN in occluded face recognition, this study proposes a new PCANet deep learning network for face recognition based on L1-2DPCA. The proposed network uses L1-2DPCA to learn filters of multiple convolutional layers. After the convolutional layer, pooling is performed through binary hashing and block-by-block histogram. CNN, PCANet, 2DPCANet and L1-PCANet are compared, and the proposed network is tested on AR and RMFD face data sets. The results show that in all tests, the recognition performance of the deep learning network proposed in this study is better than other networks. Due to the use of L1 norm, the proposed deep learning network has strong robustness to the changes of outliers and training images.

Key words: face recognition, occlusion, deep learning, L1-2DPCA, two-dimensional principal component analysis, L1 norm, CNN, robustness

中图分类号:

TN432

程晓雅,张雷. 基于深度学习的有遮挡人脸识别方法研究[J]. 电子科技, 2022, 35(1): 35-39.

CHENG Xiaoya,ZHANG Lei. Research on Occluded Face Recognition Method Based on Deep Learning[J]. Electronic Science and Technology, 2022, 35(1): 35-39.

图/表 8

图1

图2

图3

表1

图4

表2

图5

表3

参考文献 15

[1]	Wang H, Qin J, Hou Q, et al. Cattle face recognition method based on parameter transfer and deep learning[J]. Journal of Physics:Conference Series, 2020, 1453(1):2054-2062.
[2]	Zhou L, Gao M, He C. Study on face recognition under unconstrained conditions based on LBP and deep learning[J]. Journal of Computational Methods in Sciences and Engineering, 2020(1):1-12.
[3]	黄超, 薄华. 基于卷积神经网络结合SVM的人脸识别研究[J]. 信息技术与网络安全, 2017, 36(15):56-58.
	Huang Chao, Bo Hua. Study on face recognition based on CNN and SVM[J]. Information Technology and Network Security, 2017, 36(15):56-58.
[4]	潘峥嵘, 王震. 一种结合Gabor小波与深度学习的人脸识别方法[J]. 计算机与现代化, 2018(11):46-50.
	Pan Zhengrong, Wang Zhen. A face recognition algorithm based on Gabor wavelet and deep learning[J]. Computer and Modernization, 2018(11):46-50.
[5]	Payal P, Goyani M M. A comprehensive study on face recognition:methods and challenges[J]. The Imaging Science Journal, 2020, 68(2):114-127. doi: 10.1080/13682199.2020.1738741
[6]	Ruan S, Tang C, Xu Z, et al. Multi-pose face recognition based on deep learning in unconstrained scene[J]. Applied Sciences, 2020, 10(13):4669-4682. doi: 10.3390/app10134669
[7]	周慧敏, 杨明. 基于MBP算法和深度学习的人脸识别[J]. 河北工业科技, 2019, 36(1):25-30.
	Zhou Huimin, Yang Ming. Face recognition based on monogenic binary patterns and deep learning[J]. Hebei Journal of Industrial Science & Technology, 2019, 36(1):33-38.
[8]	Iqbal M, Sameem M S I, Naqvi N. A deep learning approach for face recognition based on angularly discriminative features-ScienceDirect[J]. Pattern Recognition Letters, 2019, 128(3):414-419. doi: 10.1016/j.patrec.2019.10.002
[9]	Chan T H, Jia K, Gao S, et al. PCANet: a simple deep learning baseline for image classification?[J]. IEEE Transactions on Image Processing, 2015, 24(12):5017-5032. doi: 10.1109/TIP.2015.2475625
[10]	Shi J, Wu J, Li Y, et al. Histopathological image classification with color pattern random binary hashing-based PCANet and matrix-form classifier[J]. IEEE Journal of Biomedical and Health Informatics, 2016, 21(5):1327-1337. doi: 10.1109/JBHI.2016.2602823
[11]	Yan L, Hu S, Zhang A, et al. A classification method for hyperspectral imagery based on deep learning[J]. Artificial Intelligence and Robotics Research, 2017, 6(1):31-39. doi: 10.12677/AIRR.2017.61005
[12]	Moustafa A A, Elnakib A, Areed N F F. Optimization of deep learning features for age-invariant face recognition[J]. International Journal of Electrical and Computer Engineering, 2020, 10(2):1833-1856.
[13]	Lin C H, Wang Z H, Jong G J. A de-identification face recognition using extracted thermal features based on deep learning[J]. IEEE Sensors Journal, 2020, 20(16):9510-9517.
[14]	Pei Z, Xu H, Zhang Y, et al. Face recognition via deep learning using data augmentation based on orthogonal experiments[J]. Electronics, 2019, 8(10):88-97. doi: 10.3390/electronics8010088
[15]	马祥, 李文敏, 付俊妮. 基于L1范数特征的人脸表情识别算法[J]. 电子设计工程, 2018, 26(9):163-166.
	Ma Xiang, Li Wenmin, Fu Junni. Facial experssion recognition algorithm based on L1 norm and eigenface[J]. Electronic Design Engineering, 2018, 26(9):163-166.

	无遮挡	戴墨镜	戴围巾	戴墨镜和围巾
CNN	78.94±4.05	78.84±3.95	79.34±3.85	80.01±3.85
PCANet	78.72±3.15	78.80±4.51	79.45±3.97	80.40±4.10
2DPCANet	83.21±4.21	83.79±4.35	83.81±3.02	83.44±4.27
L1-PCANet	86.92±0.51	86.79±0.42	87.15±0.35	86.46±0.22
本文方法	89.35±0.39	88.78±0.32	88.90±0.45	88.52±0.19

	10%	20%	30%	50%
PCANet	92.59±0.39	87.97±0.42	75.02±0.46	45.23±0.75
2DPCANet	93.97±0.31	88.69±0.49	78.85±0.85	56.41±0.68
L1-PCANet	94.26±0.41	91.35±0.49	84.26±0.59	66.11±0.63
本文方法	98.88±0.16	98.12±0.17	95.53±0.22	83.21±0.72

	1	2	3	4	5	6	7	均值	RMSE
PCANet	76.56	77.56	76.90	68.73	74.40	70.56	79.84	74.94	4.42
2DPCANet	73.81	76.81	76.81	74.31	78.97	74.14	74.64	75.64	2.42
L1-PCANet	83.49	82.83	82.66	83.16	85.66	83.16	82.49	83.35	1.65
本文方法	86.64	85.47	86.14	87.14	87.97	87.47	87.47	86.90	1.45