三维深度点云监督和置信度修正的人脸欺诈检测算法

doi:10.12263/DZXB.20220949

电子学报

• •

三维深度点云监督和置信度修正的人脸欺诈检测算法

胡永健¹^,³, 蔡楚鑫¹^,³, 刘琲贝¹, 王宇飞², 廖广军²

^1.华南理工大学电子与信息学院，广东广州 510641
^2.广东警官学院刑事技术系，广东广州 510440
^3.中新国际联合研究院，广东广州 510700

收稿日期:2022-08-10 修回日期:2022-11-06 出版日期:2023-02-24
- 通讯作者:
- 刘琲贝
- 作者简介:
- 胡永健男，1962年出生，湖北武汉人.华南理工大学电子与信息学院教授.主要研究方向为多媒体信息安全、图像处理、人工智能及其应用.E-mail: eeyjhu@scut.edu.cn
  蔡楚鑫男，1996年出生，广东揭阳人.华南理工大学电子与信息学院硕士研究生.主要研究方向为多媒体信息安全、图像处理、人工智能及其应用.E-mail: eechuxincai@mail.scut.edu.cn
  刘琲贝女，1980年出生，广东广州人.华南理工大学电子与信息学院讲师.主要研究方向为多媒体信息安全、图像处理、人工智能及其应用.E-mail: eebbliu@scut.edu.cn
  王宇飞男，1987年出生，海南海口人.广东警官学院刑事技术系讲师.主要研究方向为多媒体信息安全、图像处理、人工智能及其应用.
  廖广军男，1981 年出生，四川渠县人 . 广东警官学院刑事技术系教授 . 主要研究方向为多媒体信息安全.E-mail: 20060266@gdppla.edu.cn
- 基金资助:
- 国家重点研发计划(2019QY2202);广州黄埔开发区国际合作项目(2019GH16);2021年度广东省重点建设学科科研能力提升项目(2021ZDJS047);教育部科技部司法鉴定技术应用与社会治理学科创新引智基地项目(B20077)

3D Depth Point Cloud Supervision and Confidence Correction for Face Spoofing Detection

HU Yong-Jian¹^,³, CAI Chu-Xin¹^,³, LIU Bei-Bei¹, WANG Yu-Fei², LIAO Guang-Jun²

^1.School of Electronic and Information Engineering，South China University of Technology，Guangzhou，Guangdong 510641，China
^2.Department of Criminal Science and Technology，Guangdong Police College，Guangzhou，Guangdong 510440，China
^3.China-Singapore International Joint Research Institute，Guangzhou，Guangdong 510700，China

Received:2022-08-10 Revised:2022-11-06 Online:2023-02-24
- Corresponding author:
- LIU Bei-Bei
- Supported by:
- National Key Research and Development Project(2019QY2202);Science and Technology Foundation of Guangzhou Huangpu Development District(2019GH16);2021 Scientific Research Capability Improvement Program for Key Discipline Construction of Guangdong Province(2021ZDJS047);Forensic Sciences and Social Governance Disciplinary Innovation Base(B20077)

摘要/Abstract

摘要：

基于深度学习的人脸身份认证由于使用便捷和用户体验好，成为我国当今最受欢迎的人工智能技术应用之一.人脸识别和认证系统必须确保所比对的人脸是真实人脸，否则输出的结果没有任何商业价值.位于系统前端的人脸欺诈检测也称活体检测是保障人脸识别和认证系统有效输出的关键.现有人脸欺诈检测算法虽然库内性能尚佳，但由于实验室训练环境无法完全模拟真实应用场景，造成源域和目标域的数据在分布上存在差异，导致跨库检测性能明显下降.尽管通过增加检测特征的种类和个数可以改善算法性能，但会导致检测网络构造复杂，模型变大，计算复杂度增加.为了改善算法的跨库检测性能并降低计算的复杂度，本文提出一种基于三维（3D）深度点云监督和置信度修正机制的人脸欺诈检测算法.主要贡献包括：设计了DenseBlockNet，仅用较浅层的DenseBlockNet网络即可提取真假人脸之间具有很好区分度的深度信息特征，模型小；将DenseBlockNet输出的二维深度图与采样点位置进行关联，构造三维深度点云，采用倒角损失函数监督预测的深度点云与实际点云标签之间的三维空间距离，同时还采用图二元交叉熵损失监督预测的深度图与深度图标签之间的差异；在3D深度点云预测模块中引入置信度修正机制，修正二分类误差，同时避免库内过拟合，提高算法的泛化能力.所提出方法与包括2种最新文献的8种典型算法在Replay-attack、CASIA-FASD、MSU-MFSD、Rose-Youtu、OULU-NPU等5个主流人脸欺诈检测数据库上进行了充分的对比实验，实验结果表明，所提出的算法在库内和跨库检测中均能保持半总错误率最低或次低，且模型最小，参数量最少，计算复杂度最低.

关键词: 人脸欺诈检测, 三维深度点云, 3D深度点云监督, 置信度修正, 深度学习, 泛化能力

Abstract:

Due to feasibility and friendly user interaction, deep learning-based face recognition and identity authentication becomes one of the most popular artificial intelligence technologies in China. The face recognition and identity authentication system should secure that the captured face for verification is a living face rather than a fake face or called spoofing face. Otherwise, the output of the system is useless for business. Face spoofing detection or called living face detection mechanism is set in the front part of the system, and plays a key role in distinguishing a fake face from the input faces. Most current face anti-spoofing algorithms perform well in intra-dataset. However, the model training in lab is unable to simulate all aspects in the real-world application scenarios. As a result, the data distribution in source domain is not always similar to the data distribution in target domain, which causes the lab-trained algorithms barely perform as well as in lab. Although we can mitigate the performance degradation with the increase of detection feature types and dimensions, it tends to make the detection network very complex in structure and large in model size. In order to improve the generalization ability of model without resorting to large model, we design a face spoofing detection network using the 3D depth point cloud supervision and confidence correction scheme. The proposed approach consists of three major contributions. First, we design a shallow convolutional neural network called DenseBlockNet. It can well extract distinctive depth features between real faces and spoofing ones and has a small model size. Second, we establish the relationship between the 2D depth map produced by DenseBlockNet and the coordinates of sampling points, and thus create a 3D depth point cloud. We adopt the Chamfer loss to minimize the distance between the learned 3D depth point cloud and the ground truth 3D depth point cloud label, and use the Binary Cross Entropy loss to supervise the difference between the learned 2D depth map and the ground truth 2D depth map label. Third, we introduce a prediction confidence map to correct the error of the learned 3D depth point cloud, so that it can avoid overfitting in intra-datasets and obtain good generalization ability in inter-datasets. Extensive experiments are conducted on 5 popular presentation attack databases, namely Reply-attack, CASIA-FASD, MSU-MFSD, Rose-Youtu, and OULU-NPU. Compared with 8 representative methods including 2 SOTA methods, the proposed method can achieve the least or second least half-total-error-rates in either intra-dataset or inter-dataset tests. Besides, it has the smallest model, the least amount of model parameters and the lowest computational complexity.

Key words: face spoofing detection, 3D depth point cloud, 3D point cloud supervision, confidence correction, deep learning, generalization ability

中图分类号:

TP391.4

胡永健, 蔡楚鑫, 刘琲贝, 等. 三维深度点云监督和置信度修正的人脸欺诈检测算法[J]. 电子学报, DOI: 10.12263/DZXB.20220949.

Yong-Jian HU, Chu-Xin CAI, Bei-Bei LIU, et al. 3D Depth Point Cloud Supervision and Confidence Correction for Face Spoofing Detection[J]. Acta Electronica Sinica, DOI: 10.12263/DZXB.20220949.

图/表 17

参考文献 36

1	丁嵘, 苏光大, 林行刚. 使用关键点信息改进弹性匹配人脸识别算法[J]. 电子学报, 2002, 30(9): 1292-1294.
	DING R, SU G D, LIN X G. Using key points to improve elastic matching in face recognition[J]. Acta Electronica Sinica, 2002, 30(9): 1292-1294. (in Chinese)
2	胡永健, 王宇飞, 刘琲贝, 等. 人脸欺诈检测最新进展及典型方法[J]. 信号处理, 2021, 37(12): 2261-2277.
	HU Y J, WANG Y F, LIU B B, et al. A survey on the latest development and typical methods of face anti-spoofing[J]. Journal of Signal Processing, 2021, 37(12): 2261-2277. (in Chinese)
3	BOULKENAFET Z, KOMULAINEN J, HADID A. Face spoofing detection using colour texture analysis[J]. IEEE Transactions on Information Forensics and Security, 2016,11(8):1818-1830.
4	CHEN H N, HU G S, LEI Z, et al. Attention-based two-stream convolutional networks for face spoofing detection[J]. IEEE Transactions on Information Forensics and Security, 2020, 15: 578-593.
5	YU Z T, ZHAO C X, WANG Z Z, et al. Searching central difference convolutional networks for face anti-spoofing[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 5294-5304.
6	PINTO A, GOLDENSTEIN S, FERREIRA A, et al. Leveraging shape, reflectance and albedo from shading for face presentation attack detection[J]. IEEE Transactions on Information Forensics and Security, 2020, 15: 3347-3358.
7	YU Z T, LI X B, NIU X S, et al. Face anti-spoofing with human material perception[C]//European Conference on Computer Vision. Cham: Springer, 2020: 557-575.
8	LIU S Q, LAN X Y, YUEN P C. Multi-channel remote photoplethysmography correspondence feature for 3D mask face presentation attack detection[J]. IEEE Transactions on Information Forensics and Security, 2021, 16: 2683-2696.
9	ATOUM Y, LIU Y, JOURABLOO A, et al. Face anti-spoofing using patch and depth-based CNNs[C]//2017 IEEE International Joint Conference on Biometrics (IJCB). Piscataway: IEEE, 2017: 319-328.
10	GEORGE A, MARCEL S. Deep pixel-wise binary supervision for face presentation attack detection[C]//2019 International Conference on Biometrics (ICB). Piscataway: IEEE, 2019: 1-8.
11	WANG Z Z, YU Z T, ZHAO C X, et al. Deep spatial gradient and temporal depth learning for face anti-spoofing[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 5041-5050.
12	LI X, WAN J, JIN Y, et al. 3DPC-net: 3D point cloud network for face anti-spoofing[C]//2020 IEEE International Joint Conference on Biometrics. Piscataway: IEEE, 2020: 1-8.
13	LIU W H, WEI X K, LEI T, et al. Data fusion based two-stage cascade framework for multi-modality face anti-spoofing[J]. IEEE Transactions on Cognitive and Developmental Systems, 2022, 14(2): 672-683.
14	LIU A J, TAN Z C, WAN J, et al. Face anti-spoofing via adversarial cross-modality translation[J]. IEEE Transactions on Information Forensics and Security, 2021, 16: 2759-2772.
15	SIGMUND D, KERCKHOFF F, MAGDALENO J Y, et al. Face presentation attack detection in ultraviolet spectrum via local and global features[C]//2020 International Conference of the Biometrics Special Interest Group (BIOSIG). Piscataway: IEEE, 2020: 1-5.
16	QUAN R J, WU Y, YU X, et al. Progressive transfer learning for face anti-spoofing[J]. IEEE Transactions on Image Processing, 2021, 30: 3946-3955.
17	QIN Y X, ZHAO C X, ZHU X Y, et al. Learning meta model for zero-and few-shot face anti-spoofing[C]//The 36th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI, 2020: 11916-11923.
18	CAI R Z, LI H L, WANG S Q, et al. DRL-FAS: A novel framework based on deep reinforcement learning for face anti-spoofing[J]. IEEE Transactions on Information Forensics and Security, 2020, 16: 937-951.
19	CAI R Z, LI Z, WAN R J, et al. Learning meta pattern for face anti-spoofing[J]. IEEE Transactions on Information Forensics and Security, 2022, 17: 1201-1213.
20	JOURABLOO A, LIU Y J, LIU X M. Face de-spoofing: anti-spoofing via noise modeling[C]//European Conference on Computer Vision. Cham: Springer, 2018: 297-315.
21	ZHANG K Y, YAO T P, ZHANG J, et al. Face anti-spoofing via disentangled representation learning[C]//European Conference on Computer Vision. Cham: Springer, 2020: 641-657.
22	WANG G Q, HAN H, SHAN S G, et al. Cross-domain face presentation attack detection via multi-domain disentangled representation learning[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 6678-6687.
23	QIN Y X, ZHANG W G, SHI J P, et al. One-class adaptation face anti-spoofing with loss function search[J]. Neurocomputing, 2020, 417: 384-395.
24	FATEMIFAR S, ARASHLOO S R, AWAIS M, et al. Client-specific anomaly detection for face presentation attack detection[J]. Pattern Recognition, 2021, 112: 107696.
25	LI Z, LI H L, LAM K Y, et al. Unseen face presentation attack detection with hypersphere loss[C]//2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway: IEEE, 2020: 2852-2856.
26	ZHU X Y, LIU X M, LEI Z, et al. Face alignment in full pose range: A 3D total solution[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(1): 78-92.
27	SHELHAMER E, LONG J, DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(4): 640-651.
28	Guo C, Pleiss G, Sun Y, et al. On calibration of modern neural networks[C]//International Conference on Machine Learning. Cambridge MA: JMLR, 2017: 1321-1330.
29	KENDALL A, GAL Y. What uncertainties do we need in Bayesian deep learning for computer vision?[C]//The 31st International Conference on Neural Information Processing Systems. New York: Curran Associates Inc., 2017: 5580-5590.
30	CHINGOVSKA I, ANJOS A, MARCEL S. On the effectiveness of local binary patterns in face anti-spoofing[C]//2012 International Conference of Biometrics Special Interest Group (BIOSIG). Piscataway: IEEE, 2012: 1-7.
31	ZHANG Z W, YAN J J, LIU S F, et al. A face antispoofing database with diverse attacks[C]//The 5th IAPR International Conference on Biometrics (ICB). Piscataway: IEEE, 2012: 26-31.
32	WEN D, HAN H, JAIN A K. Face spoof detection with image distortion analysis[J]. IEEE Transactions on Information Forensics and Security, 2015, 10(4): 746-761.
33	LI H L, LI W, CAO H, et al. Unsupervised domain adaptation for face anti-spoofing[J]. IEEE Transactions on Information Forensics and Security, 2018, 13(7): 1794-1809.
34	BOULKENAFET Z, KOMULAINEN J, LI L, et al. OULU-NPU: A mobile face presentation attack database with real-world variations[C]//The 12th IEEE International Conference on Automatic Face & Gesture Recognition. Piscataway: IEEE, 2017: 612-618.
35	LI Z, CAI R Z, LI H L, et al. One-class knowledge distillation for face presentation attack detection[J]. IEEE Transactions on Information Forensics and Security, 2022, 17: 2137-2150.
36	WU H T, ZENG D, HU Y B, et al. Dual spoof disentanglement generation for face anti-spoofing with depth uncertainty learning[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2022, 32(7): 4626-4638.

	网络层	输入通道数(个)	步长(像素)	填充值(像素)	输出尺寸(像素)	输出通道数(个)
	输入	3	-	-	224×224	3
初步提取	7×7卷积	3	2	3	112×112	64
初步提取	最大池化	64	2	1	56×56	64
DenseLayer_1	1×1卷积	64	1	0	56×56	128
DenseLayer_1	3×3卷积	128	1	1	56×56	32
DenseLayer_2	1×1卷积	96	1	0	56×56	128
DenseLayer_2	3×3卷积	128	1	1	56×56	32
DenseLayer_3	1×1卷积	128	1	0	56×56	128
DenseLayer_3	3×3卷积	128	1	1	56×56	32
DenseLayer_4	1×1卷积	160	1	0	56×56	128
DenseLayer_4	3×3卷积	128	1	1	56×56	32
DenseLayer_5	1×1卷积	192	1	0	56×56	128
DenseLayer_5	3×3卷积	128	1	1	56×56	32
DenseLayer_6	1×1卷积	224	1	0	56×56	128
DenseLayer_6	3×3卷积	128	1	1	56×56	32
TransitionBlock	1×1卷积	256	1	0	56×56	128
TransitionBlock	平均池化	128	2	0	28×28	128
	3×3卷积	128	1	1	28×28	256
	平均池化	256	2	0	14×14	256

	网络层	输入通道数(个)	步长(像素)	填充值(像素)	输出尺寸(像素)	输出通道数(个)
	输入	3	-	-	224×224	3
初步提取	7×7卷积	3	2	3	112×112	64
初步提取	最大池化	64	2	1	56×56	64
DenseLayer_1	1×1卷积	64	1	0	56×56	128
DenseLayer_1	3×3卷积	128	1	1	56×56	32
DenseLayer_2	1×1卷积	96	1	0	56×56	128
DenseLayer_2	3×3卷积	128	1	1	56×56	32
DenseLayer_3	1×1卷积	128	1	0	56×56	128
DenseLayer_3	3×3卷积	128	1	1	56×56	32
DenseLayer_4	1×1卷积	160	1	0	56×56	128
DenseLayer_4	3×3卷积	128	1	1	56×56	32
DenseLayer_5	1×1卷积	192	1	0	56×56	128
DenseLayer_5	3×3卷积	128	1	1	56×56	32
DenseLayer_6	1×1卷积	224	1	0	56×56	128
DenseLayer_6	3×3卷积	128	1	1	56×56	32
TransitionBlock	1×1卷积	256	1	0	56×56	128
TransitionBlock	平均池化	128	2	0	28×28	128
	3×3卷积	128	1	1	28×28	256
	平均池化	256	2	0	14×14	256

实验参数	取值
网络输入尺寸	224×224×3
深度阈值T_d	0.7
损失权重λ₁	1.5
损失权重λ₂	1.0
损失权重λ₃	0.5
损失权重λ₄	0.1
批量尺寸Batch Size	32
SGD优化初始学习率	1×10^-3
学习率衰减触发周期	10个
学习率衰减倍数	0.8
最大训练周期	300个

实验参数	取值
网络输入尺寸	224×224×3
深度阈值T_d	0.7
损失权重λ₁	1.5
损失权重λ₂	1.0
损失权重λ₃	0.5
损失权重λ₄	0.1
批量尺寸Batch Size	32
SGD优化初始学习率	1×10^-3
学习率衰减触发周期	10个
学习率衰减倍数	0.8
最大训练周期	300个

训练库→测试库	C→C	R→R	M→M	Y→Y
指标(%)	EER	EER	EER	EER
Colour Texture^[3]	2.28	2.10	4.93	7.07
MSR-MobileNet^[4]	3.66	0.43	3.03	5.48
DRL-FAS^[18]	0.20	0.00	0.26	1.80
DeepPixBis^[10]	1.46	0.01	0.33	1.40
3DPC-Net^[12]	1.77	0.04	0.98	2.70
OCDA^[35]	4.27	0.47	0.68	3.96
DSDG^[36]	1.22	0.44	0.42	1.62
本文方法	0.55	0.00	0.01	1.21

三维深度点云监督和置信度修正的人脸欺诈检测算法

3D Depth Point Cloud Supervision and Confidence Correction for Face Spoofing Detection

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 17

参考文献 36

相关文章 15

编辑推荐

Metrics

测试策略	P1	P2	P3	P4
指标(%)	ACER	ACER	ACER	ACER
MSR-MobileNet^[4]	6.70	6.30	6.30±2.20	11.30±3.90
DRL-FAS^[18]	4.70	1.90	3.00±1.50	7.20±3.90
DeepPixBis^[10]	0.42	5.97	11.11±9.40	25.00±12.67
DepthMap^[11]	1.00	1.90	2.70±0.60	5.00±2.20
3DPC-Net^[12]	1.20	3.00	2.80±0.50	3.50±5.40
DeSpoofing^[20]	1.50	4.30	3.60±1.60	5.60±5.70
OCDA^[35]	4.51	5.57	4.95±1.41	12.40±5.44
DSDG^[36]	0.37	1.37	1.40±1.50	2.30±2.30
本文方法	0.23	1.41	0.88±0.65	1.38±1.03

训练库→测试库	C→R	R→C	M→R	R→M	Y→C	Y→R
指标(%)	HTER	HTER	HTER	HTER	HTER	HTER
Colour Texture^[3]	29.80	38.16	35.20	33.47	41.35	28.11
MSR-MobileNet^[4]	30.25	33.53	26.48	38.55	32.45	20.86
DRL-FAS^[18]	28.40	33.20	29.70	15.60	8.10	20.00
DeepPixBis^[10]	25.71	35.22	27.42	38.54	25.72	22.00
3DPC-Net^[12]	22.26	27.73	27.60	20.91	17.41	21.16
DeSpoofing^[20]	28.50	41.10	33.20	27.80	37.20	38.50
OCDA^[35]	3.50	31.90	2.90	20.80	21.70	3.00
DSDG^[36]	15.10	26.70	21.90	17.43	12.51	17.43
本文方法	12.58	20.34	6.36	13.56	10.80	9.17

训练库	C库		R库			M库		Y库
测试库	C(库内)	R(跨库)	R(库内)	C(跨库)	M(跨库)	M(库内)	R(跨库)	Y(库内)	C(跨库)	R(跨库)
DeepPixBis-Dense	0.75	17.11	0.02	22.71	17.16	0.17	12.72	1.32	11.46	10.80
DenseBlockNet	0.63	15.24	0.00	21.68	15.30	0.01	7.61	1.27	11.28	10.19

网络类型	网络结构	参数量(M)	FLOPs (G)	模型大小(MB)
基本网络	ResNet-18	11.68	2.38	44.67
	Inception-V3	27.16	2.85	103.94
	DenseNet169	14.15	3.42	54.96
图监督	DeepPixBis^[10]	1.43	2.11	5.60
3D点云监督	3DPC-Net^[12]	11.34	1.91	43.34
3D点云监督	本文方法	0.74	1.53	2.89

[1]	李钦, 刘伟, 牛朝阳, 宝音图, 惠周勃. 低信噪比下基于分裂EfficientNet网络的雷达信号调制方式识别[J]. 电子学报, 2023, 51(3): 675-686.
[2]	范兵兵, 何庭建, 张聪炫, 陈震, 黎明. 联合遮挡约束与残差补偿的特征金字塔光流计算方法[J]. 电子学报, 2023, 51(3): 648-657.
[3]	张聿远, 闫文君, 张立民. 基于多模态特征融合网络的空时分组码识别算法[J]. 电子学报, 2023, 51(2): 489-498.
[4]	许新征, 李杉. 基于特征膨胀卷积模块的轻量化技术研究[J]. 电子学报, 2023, 51(2): 355-364.
[5]	李滔, 董秀成, 林宏伟. 基于深监督跨尺度注意力网络的深度图像超分辨率重建[J]. 电子学报, 2023, 51(1): 128-138.
[6]	郭晓轩, 冯其波, 冀振燕, 郑发家, 杨燕燕. 多线激光光条图像缺陷分割模型研究[J]. 电子学报, 2023, 51(1): 172-179.
[7]	贾童瑶, 卓力, 李嘉锋, 张菁. 基于深度学习的单幅图像去雾研究进展[J]. 电子学报, 2023, 51(1): 231-245.
[8]	何滢婕, 刘月峰, 边浩东, 郭威, 张小燕. 基于Informer的电池荷电状态估算及其稀疏优化方法[J]. 电子学报, 2023, 51(1): 50-56.
[9]	张永梅, 孙捷. 基于动静态特征双输入神经网络的咳嗽声诊断COVID-19算法[J]. 电子学报, 2023, 51(1): 202-212.
[10]	袁海英, 成君鹏, 曾智勇, 武延瑞. Mobile_BLNet：基于Big-Little Net的轻量级卷积神经网络优化设计[J]. 电子学报, 2023, 51(1): 180-191.
[11]	王神龙, 雍宇, 吴晨睿. 基于伪孪生神经网络的低纹理工业零件6D位姿估计[J]. 电子学报, 2023, 51(1): 192-201.
[12]	吴靖, 叶晓晶, 黄峰, 陈丽琼, 王志锋, 刘文犀. 基于深度学习的单帧图像超分辨率重建综述[J]. 电子学报, 2022, 50(9): 2265-2294.
[13]	李雪莹, 王田路, 梁鹏, 王翀. 基于系统模型的用户评论中非功能需求的自动分类[J]. 电子学报, 2022, 50(9): 2079-2089.
[14]	琚长瑞, 秦晓燕, 袁广林, 李豪, 朱虹. 尺度敏感损失与特征融合的快速小目标检测方法[J]. 电子学报, 2022, 50(9): 2119-2126.
[15]	张志昌, 于沛霖, 庞雅丽, 朱林, 曾扬扬. SMGN：用于对话状态跟踪的状态记忆图网络[J]. 电子学报, 2022, 50(8): 1851-1858.

测试策略	P1	P2	P3	P4
指标(%)	ACER	ACER	ACER	ACER
DeepPixBis-Dense	0.69	1.92	2.31±1.04	4.13±2.53
DenseBlockNet	0.35	1.73	1.21±0.68	2.80±1.84

测试策略	P1	P2	P3	P4
指标(%)	ACER	ACER	ACER	ACER
DeepPixBis-Dense	0.69	1.92	2.31±1.04	4.13±2.53
DenseBlockNet	0.35	1.73	1.21±0.68	2.80±1.84