弱对齐的跨光谱人脸检测

闫梦凯; 钱建军; 杨健

doi:10.16383/j.aas.c210058

弱对齐的跨光谱人脸检测

doi: 10.16383/j.aas.c210058

闫梦凯^{1, 2, 3,},
钱建军^{1, 2, 3,},
杨健^{1, 2, 3,}

1.
南京理工大学计算机科学与工程学院南京 210094
2.
高维信息智能感知与系统教育部重点实验室南京 210094
3.
江苏省社会安全图像与视频理解重点实验室南京 210094

基金项目: 国家自然科学基金(61876083), 国家自然科学基金联合基金(U-1713208)资助

详细信息

作者简介:
闫梦凯：南京理工大学计算机科学与工程学院博士研究生. 主要研究方向为生物生理信息测量和计算机视觉. E-mail: ymk@njust.edu.cn

钱建军：南京理工大学计算机科学与工程学院副教授. 2014年获南京理工大学博士学位. 主要研究方向为模式识别和计算机视觉. 本文通信作者. E-mail: csjqian@njust.edu.cn

杨健：南京理工大学计算机科学与工程学院教授. 2002年获南京理工大学博士学位. 主要研究方向为模式识别, 计算机视觉和机器学习. E-mail: csjyang@njust.edu.cn

计量
- 文章访问数: 1413
- HTML全文浏览量: 355
- PDF下载量: 224
- 被引次数: 0
出版历程
- 收稿日期: 2021-01-20
- 网络出版日期: 2021-07-19
- 刊出日期: 2023-01-07

Weakly Aligned Cross-spectral Face Detection

YAN Meng-Kai^{1, 2, 3
,},
QIAN Jian-Jun^{1, 2, 3
,},
YANG Jian^{1, 2, 3
,}

1.
School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094
2.
Key Laboratory of Intelligent Perception and Systems for High-dimensional Information of Ministry of Education, Nanjing 210094
3.
Jiangsu Key Laboratory of Image and Video Understanding for Social Security, Nanjing 210094

Funds: Supported by National Natural Science Foundation of China (61876083) and Joint Fund of National Natural Science Foundation of China (U1713208)

More Information

Author Bio:
YAN Meng-Kai　Ph.D. candidate at the School of Computer Science and Engineering, Nanjing University of Science and Technology. His research interest covers biophysiological information measurement and computer vision

QIAN Jian-Jun　Associate professor at the School of Computer Science and Engineering, Nanjing University of Science and Technology. He received his Ph.D. degree from Nanjing University of Science and Technology in 2014. His research interest covers pattern recognition and computer vision. Corresponding author of this paper

YANG Jian　Professor at the School of Computer Science and En-gineering, Nanjing University of Science and Technology. He received his Ph.D. degree from Nanjing University of Science and Technology in 2002. His research interest covers pattern recognition, computer vision and machine learning

摘要

摘要: 跨光谱人脸检测在活体人脸识别、体温筛查等领域有着重要的应用价值. 众所周知, 可见光人脸易于检测, 然而红外人脸难于检测, 因此借助可见光图像的人脸检测结果进而完成红外人脸检测是一种有效的解决方案. 但是跨光谱图像之间不可避免的存在偏差, 导致检测精度不高. 为了解决这一问题, 提出了一种弱对齐跨光谱图像的人脸检测算法, 该方法基于跨光谱图像之间的偏差设计了候选框布置策略, 并在此基础上提出了跨光谱特征表示方法用于选取最优候选框. 此外, 本文还构建了一个跨光谱人脸数据集. 最后, 在跨光谱人脸数据集和OTCBVS人脸数据集上的实验结果证明, 该方法能够较好地完成红外图像人脸检测任务.
- 弱对齐 /
- 跨光谱 /
- 人脸检测 /
- 计算机视觉
Abstract: Cross-spectral face detection has important application value in the field of face spoof detection and body temperature screening. It is known that it is easy to detect faces in visible light images. However, it is a challenge problem to localize faces accurately in infrared images. Therefore, it is an effective solution to complete infrared face detection based on the face detection results of visible light images. The result of above scheme is nevertheless unsatisfactory since there are unavoidable deviations between cross-spectral images. To solve this problem, this paper proposes a novel face detection algorithm based on weakly aligns cross-spectral images. This method designs the dispatch scheme of candidate box based on the deviation between cross-spectral images, and thus proposes a cross-spectral feature representation method to select the best candidate box. In addition, this paper also constructed a cross-spectral face dataset. Finally, the experimental results on the cross-spectral face and OTCBVS face datasets demontrate that the proposed method can complete the task of infrared image face detection well over the competed methods.
- Weakly aligned /
- cross-spectral /
- face detection /
- computer vision

HTML全文

图 1 跨光谱人脸检测

Fig. 1 Cross-spectral face detection

下载: 全尺寸图片幻灯片

图 2 双相机与空间内任意一点的关系

Fig. 2 The relationship between dual cameras and any point in space

下载: 全尺寸图片幻灯片

图 3 空间中任意一点在相机中的成像坐标

Fig. 3 The imaging coordinates of any point in space in the camera

下载: 全尺寸图片幻灯片

图 4 像素坐标系与图像坐标系的关系

Fig. 4 The relationship between pixel coordinate system and image coordinate system

下载: 全尺寸图片幻灯片

图 5 不同深度下的跨光谱人脸图像

Fig. 5 Cross-spectral face images at different depths

下载: 全尺寸图片幻灯片

图 6 含有运动目标的跨光谱人脸图像

Fig. 6 Cross-spectral face images with moving target

下载: 全尺寸图片幻灯片

图 7 人脸高度与其成像高度的关系

Fig. 7 Relationship between face height and image height

下载: 全尺寸图片幻灯片

图 8 跨光谱人脸检测框架

Fig. 8 Cross-spectral face detection framework

下载: 全尺寸图片幻灯片

图 9 跨光谱特征表示网络

Fig. 9 Cross-spectral feature representation network

下载: 全尺寸图片幻灯片

图 10 跨光谱特征表示网络训练方式

Fig. 10 Cross-spectral feature representation network training method

下载: 全尺寸图片幻灯片

图 11 含有部分人脸的负样本

Fig. 11 Improved negative sample selection method

下载: 全尺寸图片幻灯片

图 12 相机安装位置

Fig. 12 Camera installation location

下载: 全尺寸图片幻灯片

图 13 不同采集条件下的图像

Fig. 13 Images under different acquisition conditions

下载: 全尺寸图片幻灯片

图 14 检测结果对比图

Fig. 14 Comparison of face detection results

下载: 全尺寸图片幻灯片

表 1 测试集为CSF-白天的实验结果

Table 1 Experiment results on CSF-day

算法 IoU > 0.5 时 AP (%) IoU > 0.3 时 AP (%)

坐标映射 44.6 88.4
粗略纠正 55.9 87.9
本文算法 87.5 89.6

下载: 导出CSV

表 2 测试集为CSF-夜间的实验结果

Table 2 Experiment results on CSF-night

算法 IoU > 0.5 时 AP (%) IoU > 0.3 时 AP (%)

坐标映射 36.9 82.7
粗略纠正 50.8 82.4
本文算法 81.8 84.1

下载: 导出CSV

表 3 测试集为OTCBVS的实验结果

Table 3 Experiment results on OTCBVS

算法 IoU > 0.5 时 AP (%) IoU > 0.3 时 AP (%)

坐标映射 16.4 46.8
粗略纠正 54.5 76.5
本文算法 74.4 86.6

下载: 导出CSV

表 4 候选框召回率(%)

Table 4 Proposal recall (%)

数据集 IoU > 0.5 IoU > 0.3

CSF 96.9 98.4
OTCBVS 89.3 91.6

下载: 导出CSV

表 5 CSF中候选框的选取对模型的影响

Table 5 Influence of the selection of the proposal on the model in CSF

候选框 IoU > 0.5 时 AP (%) 时间 (ms)

1/8 71.4 9
1/8, 2/8 84.8 16
1/8, 2/8, 3/8 86.3 23
1/8, $\cdots , 4/8$ 86.4 28

下载: 导出CSV

表 6 OTCBVS中候选框的选取对模型的影响

Table 6 Influence of the selection of the proposal on the model in OTCBVS

候选框 IoU > 0.5 时 AP (%) 时间 (ms)

1/8 70.5 10
1/8, 2/8 73.1 16
1/8, 2/8, 3/8 74.2 24
1/8, $\cdots, 4/8$ 74.4 30

下载: 导出CSV

表 7 负样本类型对模型精度的影响

Table 7 Effect of negative sample type on model accuracy

负样本类型 IoU > 0.5 时 AP (%)

0 46.1
4/8 70.5
5/8 69.7
6/8 47.1
7/8 21.5
0, 4/8, 5/8, 6/8, 7/8 86.4

下载: 导出CSV

表 8 CSF数据集上的对比实验结果

Table 8 Comparative experiment results on CSF dataset

算法 IoU > 0.5 时 AP (%) IoU > 0.3 时 AP (%)

FaceBoxes 9.1 9.1
S3FD 9.1 9.1
Pyramidbox 35.9 36.3
DSFD 35.8 36.3
Tinyface 56.5 82.4
S3FD-IR 72.3 73.4
DSFD-IR 81.9 83.7
DSFD-本文算法 86.4 88.5

下载: 导出CSV

表 9 OTCBVS数据集上的对比实验结果

Table 9 Comparative experiment results on OTCBVS dataset

算法 IoU > 0.5 时 AP (%) IoU > 0.3 时 AP (%)

FaceBoxes — —
S3FD 9.1 9.1
Pyramidbox 36.1 36.1
DSFD 27.2 27.2
Tinyface 25.1 38.6
S3FD-IR 60.8 73.6
DSFD-IR 69.4 70.4
DSFD-本文算法 75.0 86.3

下载: 导出CSV

参考文献(34)

[1]	Bilodeau G A, Torabi A, St-Charles P L, Riahi D. Thermal–visible registration of human silhouettes: A similarity measure performance evaluation. Infrared Physics & Technology, 2014, 64: 79-86.
[2]	Rowley H A, Baluja S, Kanade T. Neural network-based face detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1998, 20(1), 23-38. doi: 10.1109/34.655647
[3]	Rowley H A, Baluja S, Kanade T. Rotation invariant neural network-based face detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Santa Barbara, USA: IEEE, 1998. 38−44
[4]	Viola P, Jones M J. Robust Real-Time Face Detection. International Journal of Computer Vision, 2004, 57(2): 137-154. doi: 10.1023/B:VISI.0000013087.49260.fb
[5]	Li S Z, Zhu L, Zhang Z, Blake A, Zhang H, Shum H. Statistical learning of multi-view face detection. In: Proceedings of the 7th European Conference on Computer Vision. Berlin, Germany: 2002. 67−81
[6]	Mathias M, Benenson R, Pedersoli M, Van Gool L. Face detection without bells and whistles. In: Proceedings of the 13rd European Conference on Computer Vision. Zurich, Switzerland: 2014. 720−735
[7]	Li Q, Sun Z, He R, Tan T. Learning symmetry features for face detection based on sparse group lasso. In: Proceedings of the Chinese Conference on Biometric Recognition. Jinan, China: 2013. 162−169
[8]	Li J, Wang Y, Wang C, Tai Y, Qian J, Yang J, et al. DSFD: Dual shot face detector. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE, 2020. 5055−5064
[9]	Zhang S, Zhu X, Lei Z, Shi H, Wang X, Li S Z, et al. S³FD: Single shot scale-invariant face detector. In: Proceedings of the IEEE International Conference on Computer Vision. Venice, Ita-ly: IEEE, 2017. 192−201
[10]	Bai Y, Zhang Y, Ding M, Ghanem B. Finding tiny faces in the wild with generative adversarial network. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 21–30
[11]	Zhang S, Wen L, Shi H, Lei Z, Lyu S, Li S Z. Single-Shot Scale-Aware Network for Real-Time Face Detection. International Journal of Computer Vision, 2019, 127(6-7): 537-559. doi: 10.1007/s11263-019-01159-3
[12]	Zhang S, Zhu X, Lei Z, Shi H, Wang X, Li S Z. Faceboxes: A CPU real-time face detector with high accuracy. In: Proceedin-gs of the IEEE International Joint Conference on Biometrics. Denver, USA: IEEE, 2017. 1−9
[13]	Tang X, Du D K, He Z, Liu J. Pyramidbox: A context-assisted single shot face detector. In: Proceedings of the 15th European Conference on Computer Vision. Munich, Germany: 2018. 797− 813
[14]	Yang S, Luo P, Loy C C, Tang X. Wider face: A face detection benchmark. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE, 2016. 525−5533
[15]	C. Ma, T Ngo, U Hideaki, N Hajime, S. Atsushi, T Rin-Ichiro. Adapting local features for face detection in thermal image. Sensors, 2017, 17(12): 2741-.
[16]	Kyal C K, Poddar H, Reza M. Detection of human face by thermal infrared camera using MPI model and feature extraction method. In: Proceedings of the 4th International Conference on Computing Communication and Automation. Greater Noida, India: IEEE, 2018. 1−5
[17]	Budzan S, Wyżgolik R. Face and eyes localization algorithm in thermal images for temperature measurement of the inner canthus of the eyes. Infrared Physics & Technology, 2013, 60: 225-234.
[18]	Ribeiro R F, Fernandes J M, Neves A J R. Face detection on infrared thermal image. In: Proceedings of the 2nd International Conference on Advances in Signal Image and Video Processing. Barcelona, Spain: 2017. 38–42
[19]	Goulart C, Valadão C, Delisle-Rodriguez D, Funayama D, Favarato A, Baldo G, et al. Visual and Thermal Image Processing for Facial Specific Landmark Detection to Infer Emotions in a Child-Robot Interaction. Sensors, 2019, 19(13): 2844. doi: 10.3390/s19132844
[20]	Somboonkaew A, Prempree P, Vuttivong S, Wetcharungsri J, Porntheeraphat S, Chanhorm S, et al. Mobile-platform for automatic fever screening system based on infrared forehead temperature. In: Proceedings of the IEEE Opto-Electronics and Communications Conference and Photonics Global Conference. Sing-apore, Singapore: IEEE, 2017. 1−4
[21]	Mallat K, Dugelay J L. A benchmark database of visible and thermal paired face images across multiple variations. In: Proceedings of the IEEE International Conference of the Biometrics Special Interest Group. Darmstadt, Germany: IEEE, 2018. 1−5
[22]	Panetta K, Wan Q, Agaian S, Rajeev S, Kamath S, Rajendran R, et al. A comprehensive database for benchmarking imaging systems. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 42(3): 509-520.
[23]	Hwang S, Park J, Kim N, Choi Y, So Kweon I. Multispectral pedestrian detection: Benchmark dataset and baseline. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE, 2015. 1037−1045
[24]	张秀伟, 张艳宁, 杨涛, 等. 基于Co-motion的可见光--热红外图像序列自动配准算. 自动化学报, 2010, 36(009): 1220-1231. doi: 10.3724/SP.J.1004.2010.01220 Zhang Xiu-Wei, Zhang Yan-Ning, Yang Tao, Zhang Xin-Gong, Shao Da-Pei. Automatic Visual-thermal Image Sequence Registration Based on Co-motion. Acta Automatica Sinica, 2010, 36(009): 1220-1231. doi: 10.3724/SP.J.1004.2010.01220
[25]	高雪琴, 刘刚, 肖刚, 等. 基于FPDE的红外与可见光图像融合算法[J]. 自动化学报, 2020, v. 46(04): 186-194. Gao Xue-Qin, Liu Gang, Xiao Gang, Bavirisetti Durga Prasad, SHI Kai-Lei. Fusion Algorithm of Infrared and Visible Images Based on FPDE. Acta Automatica Sinica, 2020, v. 46(04): 186-194.
[26]	廉蔺, 李国辉, 张军, 涂丹. 基于边缘最优映射的红外和可见光图像自动配准算. 自动化学报, 2012, 38(04): 570-581. doi: 10.3724/SP.J.1004.2012.00570 Lian Lin, Li Guo-Hui, Zhang Jun, Tu Dan. An Automatic Registration Algorithm of Infrared and Visible Images Based on Optimal Mapping of Edges. Acta Automatica Sinica, 2012, 38(04): 570-581. doi: 10.3724/SP.J.1004.2012.00570
[27]	(刘松涛, 刘振兴, 姜宁. 基于融合显著图和高效子窗口搜索的红外目标分. 自动化学报, 2018, 44(012): 2210-2221). Liu Song-Tao, Liu Zhen-Xing, Jiang Ning. Target Segmentation of Infrared Image Using Fused Saliency Map and Efficient Subwindow Search. Acta Automatica Sinica, 2018, 44(012): 2210-2221.
[28]	(袁浩期, 李扬, 王俊影, 等. 基于红外热像的行人面部温度高精度检测技术. 红外技术, 2019, v. 41;No. 324(12): 94-99). Yuan Hao-qi, Li Yang, Wang Jun-Ying, Liu Hang. High Precision Detection Technology of Pedestrian Face Temperature Based on Infrared Thermal Imaging. Infrared Technology, 2019, v. 41;No. 324(12): 94-99.
[29]	Zhi T, Pires B R, Hebert M, Narasimhan S. G. Deep material-aware cross-spectral stereo matching. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 1916−1925
[30]	Liang M, Guo X, Li H, Wang X, Song, Y. Unsupervised cross-spectral stereo matching by learning to synthesize. In: Proceedings of the AAAI Conference on Artificial Intelligence. Honolulu, USA: 2019. 8706−8713
[31]	Zhang L, Zhu X, Chen X, Yang X, Lei Z, Liu Z. Weakly aligned cross-modal learning for multispectral pedestrian detection. In: Proceedings of the IEEE International Conference on Computer Vision. Seoul, South Korea: IEEE, 2019. 5127−5137
[32]	Dwith Chenna Y N, Ghassemi P, Pfefer T J, Casamento J, Wang Q. Free-Form Deformation Approach for Registration of Visible and Infrared Facial Images in Fever Screening. Sensors, 2018, 18(2): 125. doi: 10.3390/s18010125
[33]	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. arXiv preprint, 2014, arXiv: 1409.1556
[34]	Schroff F, Kalenichenko D, Philbin J. FaceNet: A unified embedding for face recognition and clustering. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE, 2015. 815−823

施引文献

资源附件(0)

访问统计

图(14) / 表(9)

计量

文章访问数: 1413
HTML全文浏览量: 355
PDF下载量: 224
被引次数: 0

姓名
邮箱
手机号码
标题
留言内容
验证码

留言板

弱对齐的跨光谱人脸检测

doi: 10.16383/j.aas.c210058

计量

Weakly Aligned Cross-spectral Face Detection

计量

目录

算法	IoU > 0.5 时 AP (%)	IoU > 0.3 时 AP (%)
坐标映射	44.6	88.4
粗略纠正	55.9	87.9
本文算法	87.5	89.6

候选框	IoU > 0.5 时 AP (%)	时间 (ms)
1/8	71.4	9
1/8, 2/8	84.8	16
1/8, 2/8, 3/8	86.3	23
1/8, $\cdots , 4/8$	86.4	28

负样本类型	IoU > 0.5 时 AP (%)
0	46.1
4/8	70.5
5/8	69.7
6/8	47.1
7/8	21.5
0, 4/8, 5/8, 6/8, 7/8	86.4

数据集	IoU > 0.5	IoU > 0.3
CSF	96.9	98.4
OTCBVS	89.3	91.6

算法	IoU > 0.5 时 AP (%)	IoU > 0.3 时 AP (%)
FaceBoxes	9.1	9.1
S3FD	9.1	9.1
Pyramidbox	35.9	36.3
DSFD	35.8	36.3
Tinyface	56.5	82.4
S3FD-IR	72.3	73.4
DSFD-IR	81.9	83.7
DSFD-本文算法	86.4	88.5

算法	IoU > 0.5 时 AP (%)	IoU > 0.3 时 AP (%)
FaceBoxes	—	—
S3FD	9.1	9.1
Pyramidbox	36.1	36.1
DSFD	27.2	27.2
Tinyface	25.1	38.6
S3FD-IR	60.8	73.6
DSFD-IR	69.4	70.4
DSFD-本文算法	75.0	86.3

留言板

弱对齐的跨光谱人脸检测

doi: 10.16383/j.aas.c210058

计量

出版历程

Weakly Aligned Cross-spectral Face Detection

计量

出版历程

目录