电子学报 ›› 2021, Vol. 49 ›› Issue (12): 2437-2448.DOI: 10.12263/DZXB.20200599
汪宇玲1, 陈立1, 黎明2, 钟国韵1, 何月顺1, 常玉祥1, 宋伟宁1
收稿日期:
2020-06-22
修回日期:
2021-08-03
出版日期:
2021-12-25
作者简介:
基金资助:
WANG Yu-ling1, CHEN Li1, LI Ming2, ZHONG Guo-yun1, HE Yue-shun1, CHANG Yu-xiang1, SONG Wei-ning1
Received:
2020-06-22
Revised:
2021-08-03
Online:
2021-12-25
Published:
2021-12-25
Supported by:
摘要:
为提高各种不同特殊场景下的模糊人脸识别精确性和鲁棒性,本文提出一种基于迹变换和旋转增量调制编码的特征提取方法.该方法将通信语音编码技术与图像变换技术相结合,首先通过迹线旋转扫描整幅图像,并对迹线上的采样信息进行增量调制编码,从而获得多角度的全局有序结构特征,然后用支持向量机对样本图像的这些特征进行训练以分类并识别图像.实验结果表明,在各种不同模糊级的低质量人脸数据库上,本文方法对不同光照变化、不同拍摄角度、不同遮挡等不同场景的人脸图像均能取得较好的识别效果,与一些传统方法相比识别性能大幅提升,相对于VGGNet和Sphereface两种先进方法在三组不同模糊度测试图像集的平均识别率分别提高2.18%和2.20%,具有更高的识别精度和更好的鲁棒性.
中图分类号:
汪宇玲, 陈立, 黎明, 钟国韵, 何月顺, 常玉祥, 宋伟宁. 基于迹变换和旋转增量调制特征的模糊人脸识别[J]. 电子学报, 2021, 49(12): 2437-2448.
WANG Yu-ling, CHEN Li, LI Ming, ZHONG Guo-yun, HE Yue-shun, CHANG Yu-xiang, SONG Wei-ning. Rotational Delta Modulation Feature and Its Application in Blurry Face Recognition Based on Trace Transform[J]. Acta Electronica Sinica, 2021, 49(12): 2437-2448.
算法 | 灰度级 | CASIA- Webface | AR | OFD | |||
---|---|---|---|---|---|---|---|
F(1) | F(2) | F(1) | F(2) | F(1) | F(2) | ||
RDM | 256 | 96.07 | 97.33 | 97.47 | 98.08 | 98.13 | 98.33 |
128 | 96.97 | 97.01 | 98.33 | 98.82 | 97.88 | 97.75 | |
64 | 96.13 | 97.33 | 97.75 | 98.58 | 97.25 | 97.75 | |
32 | 96.99 | 97.67 | 97.50 | 99.16 | 98.38 | 97.87 | |
RBDM | 256 | 96.44 | 97.67 | 96.67 | 97.10 | 96.62 | 97.25 |
128 | 96.89 | 98.71 | 97.23 | 97.35 | 97.19 | 98.62 | |
64 | 95.33 | 98.33 | 98.05 | 98.17 | 98.87 | 98.50 | |
32 | 97.56 | 98.01 | 97.70 | 98.78 | 98.25 | 98.75 |
表4 本文所提算法在正常数据集下的识别率(%)
算法 | 灰度级 | CASIA- Webface | AR | OFD | |||
---|---|---|---|---|---|---|---|
F(1) | F(2) | F(1) | F(2) | F(1) | F(2) | ||
RDM | 256 | 96.07 | 97.33 | 97.47 | 98.08 | 98.13 | 98.33 |
128 | 96.97 | 97.01 | 98.33 | 98.82 | 97.88 | 97.75 | |
64 | 96.13 | 97.33 | 97.75 | 98.58 | 97.25 | 97.75 | |
32 | 96.99 | 97.67 | 97.50 | 99.16 | 98.38 | 97.87 | |
RBDM | 256 | 96.44 | 97.67 | 96.67 | 97.10 | 96.62 | 97.25 |
128 | 96.89 | 98.71 | 97.23 | 97.35 | 97.19 | 98.62 | |
64 | 95.33 | 98.33 | 98.05 | 98.17 | 98.87 | 98.50 | |
32 | 97.56 | 98.01 | 97.70 | 98.78 | 98.25 | 98.75 |
算法 | 灰度级 | CASIA-Webface | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RDM | 256 | 94.87 | 92.40 | 91.40 | 97.89 | 96.50 | 96.67 |
128 | 93.30 | 96.10 | 91.17 | 96.89 | 95.67 | 93.83 | |
64 | 95.67 | 97.87 | 91.83 | 98.22 | 97.83 | 94.17 | |
32 | 97.87 | 97.77 | 94.40 | 97.11 | 97.00 | 95.83 | |
RBDM | 256 | 96.98 | 94.37 | 94.62 | 97.94 | 96.62 | 96.75 |
128 | 96.91 | 96.69 | 93.75 | 97.59 | 97.77 | 97.35 | |
64 | 97.86 | 96.42 | 95.14 | 98.25 | 98.10 | 97.13 | |
32 | 97.95 | 97.50 | 95.33 | 98.47 | 97.91 | 97.78 |
表5 本文所提算法在模糊数据集CASIA-Webface上的识别率(%)
算法 | 灰度级 | CASIA-Webface | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RDM | 256 | 94.87 | 92.40 | 91.40 | 97.89 | 96.50 | 96.67 |
128 | 93.30 | 96.10 | 91.17 | 96.89 | 95.67 | 93.83 | |
64 | 95.67 | 97.87 | 91.83 | 98.22 | 97.83 | 94.17 | |
32 | 97.87 | 97.77 | 94.40 | 97.11 | 97.00 | 95.83 | |
RBDM | 256 | 96.98 | 94.37 | 94.62 | 97.94 | 96.62 | 96.75 |
128 | 96.91 | 96.69 | 93.75 | 97.59 | 97.77 | 97.35 | |
64 | 97.86 | 96.42 | 95.14 | 98.25 | 98.10 | 97.13 | |
32 | 97.95 | 97.50 | 95.33 | 98.47 | 97.91 | 97.78 |
算法 | 灰度级 | AR | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RDM | 256 | 97.88 | 96.70 | 97.33 | 97.51 | 97.97 | 95.51 |
128 | 98.79 | 98.50 | 95.67 | 95.85 | 96.24 | 95.28 | |
64 | 98.25 | 98.54 | 95.33 | 97.62 | 96.72 | 94.99 | |
32 | 98.96 | 98.85 | 96.92 | 97.10 | 96.84 | 94.85 | |
RBDM | 256 | 96.67 | 94.47 | 92.56 | 97.63 | 96.89 | 96.01 |
128 | 97.23 | 97.45 | 93.25 | 97.33 | 95.67 | 94.73 | |
64 | 98.45 | 97.16 | 98.67 | 98.17 | 97.99 | 95.58 | |
32 | 98.90 | 98.96 | 94.25 | 98.78 | 97.44 | 96.67 |
表6 本文所提算法在模糊数据集AR上的识别率(%)
算法 | 灰度级 | AR | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RDM | 256 | 97.88 | 96.70 | 97.33 | 97.51 | 97.97 | 95.51 |
128 | 98.79 | 98.50 | 95.67 | 95.85 | 96.24 | 95.28 | |
64 | 98.25 | 98.54 | 95.33 | 97.62 | 96.72 | 94.99 | |
32 | 98.96 | 98.85 | 96.92 | 97.10 | 96.84 | 94.85 | |
RBDM | 256 | 96.67 | 94.47 | 92.56 | 97.63 | 96.89 | 96.01 |
128 | 97.23 | 97.45 | 93.25 | 97.33 | 95.67 | 94.73 | |
64 | 98.45 | 97.16 | 98.67 | 98.17 | 97.99 | 95.58 | |
32 | 98.90 | 98.96 | 94.25 | 98.78 | 97.44 | 96.67 |
算法 | 灰度级 | OFD | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RDM | 256 | 96.75 | 96.63 | 97.25 | 97.34 | 96.08 | 97.26 |
128 | 96.63 | 97.75 | 97.63 | 97.60 | 97.03 | 96.09 | |
64 | 98.05 | 97.13 | 97.00 | 98.28 | 98.02 | 96.03 | |
32 | 97.63 | 97.13 | 97.38 | 98.75 | 96.92 | 96.67 | |
RBDM | 256 | 96.56 | 95.62 | 95.12 | 97.87 | 96.25 | 97.71 |
128 | 97.34 | 96.04 | 95.59 | 97.56 | 96.04 | 95.12 | |
64 | 98.03 | 97.77 | 94.14 | 97.85 | 98.25 | 96.75 | |
32 | 98.65 | 98.72 | 97.07 | 98.72 | 96.47 | 97.92 |
表7 本文所提算法在模糊数据集OFD上的识别率(%)
算法 | 灰度级 | OFD | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RDM | 256 | 96.75 | 96.63 | 97.25 | 97.34 | 96.08 | 97.26 |
128 | 96.63 | 97.75 | 97.63 | 97.60 | 97.03 | 96.09 | |
64 | 98.05 | 97.13 | 97.00 | 98.28 | 98.02 | 96.03 | |
32 | 97.63 | 97.13 | 97.38 | 98.75 | 96.92 | 96.67 | |
RBDM | 256 | 96.56 | 95.62 | 95.12 | 97.87 | 96.25 | 97.71 |
128 | 97.34 | 96.04 | 95.59 | 97.56 | 96.04 | 95.12 | |
64 | 98.03 | 97.77 | 94.14 | 97.85 | 98.25 | 96.75 | |
32 | 98.65 | 98.72 | 97.07 | 98.72 | 96.47 | 97.92 |
算法 | CASIA-Webface | AR | OFD | ||||||
---|---|---|---|---|---|---|---|---|---|
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | |
LBP | 91.46 | 90.10 | 86.89 | 92.05 | 88.33 | 86.26 | 90.10 | 86.89 | 83.21 |
Trace transform[ | 90.56 | 89.66 | 85.13 | 89.91 | 87.26 | 87.19 | 91.79 | 91.46 | 89.78 |
Sphereface[ | 95.35 | 96.22 | 95.23 | 97.43 | 98.35 | 96.63 | 96.68 | 96.35 | 95.39 |
VGG19[ | 96.81 | 96.48 | 95.87 | 97.24 | 95.42 | 96.55 | 96.95 | 95.53 | 96.95 |
ResNet-101[ | 97.82 | 97.56 | 97.32 | 97.85 | 96.59 | 95.79 | 97.56 | 97.54 | 96.54 |
DenseNet[ | 97.35 | 97.36 | 97.32 | 97.59 | 97.56 | 97.17 | 97.15 | 96.32 | 96.58 |
SE-ResNeXt[ | 97.65 | 98.01 | 97.65 | 98.02 | 97.35 | 97.86 | 97.98 | 98.01 | 97.65 |
DPN[ | 97.67 | 98.02 | 98.17 | 98.89 | 98.02 | 97.82 | 98.89 | 98.25 | 98.17 |
RDM(ours) | 97.89 | 96.50 | 96.67 | 97.51 | 97.97 | 95.51 | 97.34 | 96.08 | 97.26 |
RDM+TT(ours) | 98.95 | 97.33 | 96.47 | 99.36 | 98.62 | 98.37 | 99.01 | 98.56 | 98.09 |
RBDM(ours) | 97.94 | 96.62 | 96.75 | 97.63 | 96.89 | 96.01 | 97.87 | 96.25 | 97.71 |
RBDM+TT(ours) | 98.62 | 98.12 | 98.40 | 99.04 | 98.87 | 98.59 | 99.29 | 98.23 | 98.25 |
表8 模糊数据集下本文所提算法与其他算法的识别率(%)对比
算法 | CASIA-Webface | AR | OFD | ||||||
---|---|---|---|---|---|---|---|---|---|
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | |
LBP | 91.46 | 90.10 | 86.89 | 92.05 | 88.33 | 86.26 | 90.10 | 86.89 | 83.21 |
Trace transform[ | 90.56 | 89.66 | 85.13 | 89.91 | 87.26 | 87.19 | 91.79 | 91.46 | 89.78 |
Sphereface[ | 95.35 | 96.22 | 95.23 | 97.43 | 98.35 | 96.63 | 96.68 | 96.35 | 95.39 |
VGG19[ | 96.81 | 96.48 | 95.87 | 97.24 | 95.42 | 96.55 | 96.95 | 95.53 | 96.95 |
ResNet-101[ | 97.82 | 97.56 | 97.32 | 97.85 | 96.59 | 95.79 | 97.56 | 97.54 | 96.54 |
DenseNet[ | 97.35 | 97.36 | 97.32 | 97.59 | 97.56 | 97.17 | 97.15 | 96.32 | 96.58 |
SE-ResNeXt[ | 97.65 | 98.01 | 97.65 | 98.02 | 97.35 | 97.86 | 97.98 | 98.01 | 97.65 |
DPN[ | 97.67 | 98.02 | 98.17 | 98.89 | 98.02 | 97.82 | 98.89 | 98.25 | 98.17 |
RDM(ours) | 97.89 | 96.50 | 96.67 | 97.51 | 97.97 | 95.51 | 97.34 | 96.08 | 97.26 |
RDM+TT(ours) | 98.95 | 97.33 | 96.47 | 99.36 | 98.62 | 98.37 | 99.01 | 98.56 | 98.09 |
RBDM(ours) | 97.94 | 96.62 | 96.75 | 97.63 | 96.89 | 96.01 | 97.87 | 96.25 | 97.71 |
RBDM+TT(ours) | 98.62 | 98.12 | 98.40 | 99.04 | 98.87 | 98.59 | 99.29 | 98.23 | 98.25 |
算法 | 灰 度 级 | AR | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RBDM+TT | 256 | 96.38 | 95.17 | 95.44 | 98.65 | 98.62 | 98.56 |
128 | 96.21 | 95.69 | 95.35 | 98.24 | 98.45 | 96.45 | |
64 | 98.46 | 96.42 | 95.14 | 99.02 | 99.05 | 97.23 | |
32 | 97.03 | 97.82 | 96.23 | 99.35 | 99.02 | 99.04 |
表9 RBDM+TT在不同光照强度的AR上的识别率(%)
算法 | 灰 度 级 | AR | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RBDM+TT | 256 | 96.38 | 95.17 | 95.44 | 98.65 | 98.62 | 98.56 |
128 | 96.21 | 95.69 | 95.35 | 98.24 | 98.45 | 96.45 | |
64 | 98.46 | 96.42 | 95.14 | 99.02 | 99.05 | 97.23 | |
32 | 97.03 | 97.82 | 96.23 | 99.35 | 99.02 | 99.04 |
算法 | 灰 度 级 | AR | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RBDM+TT | 256 | 97.67 | 96.37 | 96.32 | 98.45 | 97.62 | 96.53 |
128 | 97.23 | 97.69 | 96.25 | 98.19 | 97.87 | 97.85 | |
64 | 98.25 | 97.07 | 97.46 | 97.96 | 98.05 | 97.03 | |
32 | 97.17 | 97.89 | 97.33 | 98.33 | 98.21 | 97.33 |
表10 RBDM+TT在不同遮挡类型的AR上的识别率(%)
算法 | 灰 度 级 | AR | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RBDM+TT | 256 | 97.67 | 96.37 | 96.32 | 98.45 | 97.62 | 96.53 |
128 | 97.23 | 97.69 | 96.25 | 98.19 | 97.87 | 97.85 | |
64 | 98.25 | 97.07 | 97.46 | 97.96 | 98.05 | 97.03 | |
32 | 97.17 | 97.89 | 97.33 | 98.33 | 98.21 | 97.33 |
算法 | 灰 度 级 | OFD | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RBDM+TT | 256 | 97.17 | 97.33 | 97.32 | 98.96 | 98.23 | 98.25 |
128 | 97.19 | 96.39 | 97.17 | 97.32 | 97.25 | 98.24 | |
64 | 98.22 | 98.17 | 97.65 | 98.96 | 97.33 | 98.36 | |
32 | 97.96 | 98.75 | 96.96 | 99.23 | 99.17 | 98.07 |
表11 RBDM+TT在不同拍摄角度的OFD上的识别率(%)
算法 | 灰 度 级 | OFD | |||||
---|---|---|---|---|---|---|---|
F(1) | F(2) | ||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | ||
RBDM+TT | 256 | 97.17 | 97.33 | 97.32 | 98.96 | 98.23 | 98.25 |
128 | 97.19 | 96.39 | 97.17 | 97.32 | 97.25 | 98.24 | |
64 | 98.22 | 98.17 | 97.65 | 98.96 | 97.33 | 98.36 | |
32 | 97.96 | 98.75 | 96.96 | 99.23 | 99.17 | 98.07 |
算法 | AR | OFD | |||||||
---|---|---|---|---|---|---|---|---|---|
不同光照 | 不同遮挡 | 不同拍摄角度 | |||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | |
LBP | 93.05 | 89.33 | 86.76 | 91.03 | 87.23 | 86.16 | 90.17 | 86.29 | 83.31 |
Trace transform[ | 89.71 | 88.26 | 87.39 | 88.91 | 87.16 | 87.01 | 90.29 | 91.56 | 90.18 |
VGG19[ | 97.84 | 96.02 | 96.05 | 96.84 | 95.82 | 95.85 | 96.15 | 95.33 | 96.25 |
ResNet101[ | 98.25 | 96.59 | 95.72 | 97.85 | 97.59 | 95.79 | 97.26 | 97.62 | 96.34 |
DenseNet[ | 97.23 | 96.16 | 97.27 | 97.49 | 96.58 | 96.17 | 96.15 | 95.32 | 95.38 |
SE-ResNeXt[ | 98.22 | 97.65 | 97.96 | 98.02 | 97.05 | 97.46 | 96.98 | 97.01 | 97.25 |
DPN[ | 98.29 | 98.02 | 97.32 | 98.29 | 97.02 | 97.22 | 97.89 | 98.25 | 98.15 |
RBDM+TT(ours) | 99.35 | 99.05 | 99.04 | 99.45 | 98.21 | 97.85 | 99.29 | 98.23 | 98.25 |
表12 模糊数据集下RBDM+TT与其他算法的识别率(%)对比
算法 | AR | OFD | |||||||
---|---|---|---|---|---|---|---|---|---|
不同光照 | 不同遮挡 | 不同拍摄角度 | |||||||
B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | B(1) | B(2) | B(3) | |
LBP | 93.05 | 89.33 | 86.76 | 91.03 | 87.23 | 86.16 | 90.17 | 86.29 | 83.31 |
Trace transform[ | 89.71 | 88.26 | 87.39 | 88.91 | 87.16 | 87.01 | 90.29 | 91.56 | 90.18 |
VGG19[ | 97.84 | 96.02 | 96.05 | 96.84 | 95.82 | 95.85 | 96.15 | 95.33 | 96.25 |
ResNet101[ | 98.25 | 96.59 | 95.72 | 97.85 | 97.59 | 95.79 | 97.26 | 97.62 | 96.34 |
DenseNet[ | 97.23 | 96.16 | 97.27 | 97.49 | 96.58 | 96.17 | 96.15 | 95.32 | 95.38 |
SE-ResNeXt[ | 98.22 | 97.65 | 97.96 | 98.02 | 97.05 | 97.46 | 96.98 | 97.01 | 97.25 |
DPN[ | 98.29 | 98.02 | 97.32 | 98.29 | 97.02 | 97.22 | 97.89 | 98.25 | 98.15 |
RBDM+TT(ours) | 99.35 | 99.05 | 99.04 | 99.45 | 98.21 | 97.85 | 99.29 | 98.23 | 98.25 |
1 | SabourS, FrosstN, HintonG E. Dynamic routing between capsules[A]. Ulrike von Luxburg. Proceedings of the 31st International Conference on Neural Information Processing Systems[C]. New York: Curran Associates Inc, 2017. 3859 - 3869. |
2 | KimK I, KwonY. Single-image super-resolution using sparse regression and natural image prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(6): 1127 - 1133. |
3 | LiJ D, ChenZ X, LiuC Y. Low-resolution face recognition of multi-scale blocking CS-LBP and weighted PCA[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2016, 30(8):1656005:1-1656005:13. |
4 | XingX L, WangK J. Couple manifold discriminant analysis with bipartite graph embedding for low-resolution face recognition[J]. Signal Processing, 2016, 125: 329 - 335. |
5 | BiswasS, AggarwalG, FlynnP J, et al. Pose-robust recognition of low-resolution face images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(12): 3037 - 3049. |
6 | 陈松灿,伍艳莲.图像的模糊识别方法研究与实现[J]. 电子学报, 2000, 28(11): 50 - 54. |
ChenS C, WuY L. Studies and implementation of fuzzy recognition methods for image[J]. Acta Electronica Sinica, 2000, 28(11): 50 - 54. (in Chinese) | |
7 | ZafeiriouS, AtkinsonG A, HansenM F, et al. Face recognition and verification using photometric stereo: The photoface database and a comprehensive evaluation[J]. IEEE Transactions on Information Forensics and Security, 2013, 8(1): 121 - 135. |
8 | MasiI, Traâ'nA T, HassnerT, et al. Do we really need to collect millions of faces for effective face recognition?[A].European Conference on Computer Vision[C]. Amsterdam: Springer, 2016. 579 - 596. |
9 | TaigmanY, YangM, RanzatoM A, WolfL. et al. DeepFace: closing the gap to human-level performance in face verification[A]. 2014 IEEE Conference on Computer Vision and Pattern Recognition[C]. Columbus, OH, USA: IEEE Computer Society, 2014. 1701 - 1708. |
10 | SunY, WangX G, TangX O. Deep learning face representation from predicting 10,000 classes[A]. 2014 IEEE Conference on Computer Vision and Pattern Recognition[C]. Columbus, OH, USA: IEEE Computer Society, 2014. 1891 - 1898. |
11 | LiuJ T, DengY F, BaiT, et al. Targeting ultimate accuracy: Face recognition via deep embedding [EB/OL]., 2019-11-08. |
12 | SunY, WangX, TangX. Deep learning face representation by joint identification-verification[A]. Advances in Neural Information Processing Systems[C]. Montreal, Quebec, Canada: the MIT Press, 2014. 1988 - 1996. |
13 | SunY, WangX G, TangX O. Deeply learned face representations are sparse, selective, and robust[A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. Boston, USA: IEEE Computer Society, 2015. 2892 - 2900. |
14 | SchroffF, KalenichenkoD, PhilbinJ. FaceNet: A unified embedding for face recognition and clustering [A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. Boston, USA: IEEE Computer Society, 2015. 815 - 823. |
15 | LiuW Y, WenY D, YuZ D, et al. Sphereface: Deep hypersphere embedding for face recognition[A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. Hawaii, USA: IEEE Computer Society, 2017. 212 - 220. |
16 | HeinsohnD, VillalobosE, PrietoL, et al. Face recognition in low-quality images using adaptive sparse representations[J]. Image and Vision Computing, 2019, 85: 46 - 58. |
17 | WangZ F, MiaoZ J, WuQ M J, et al. Low-resolution face recognition: a review[J]. The Visual Computer, 2014, 30(4): 359 - 386. |
18 | KadyrovA, PetrouM. The trace transform and its applications[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2001, 23(8): 811 - 828. |
19 | 施展, 杜明辉, 梁亚玲. 基于2DNPP和Trace变换的平面内旋转人脸识别[J]. 华南理工大学学报(自然科学版), 2012, 40(8): 46 - 50. |
ShiZ, DuM H, LiangY L. In-plane rotary face recognition based on 2DNPP and trace transform[J]. Journal of South China University of Technology (Natural Science Edition), 2012, 40(8): 46 - 50. (in Chinese) | |
20 | WuN, LiY, TianY N, et al. Trace-transform-based time-frequency filtering for seismic signal enhancement in Northeast China[J]. Comptes Rendus Geoscience, 2016, 348(5): 360 - 367. |
21 | 汪宇玲, 黎明. 圆迹变换及图像纹理分析[J]. 电子学报, 2018, 46(10): 2351 - 2358. |
WangY L, LiM. Circular trace transform and its applications to image texture analysis[J]. Acta Electronica Sinica, 2018, 46(10): 2351 - 2358. (in Chinese) | |
22 | WangY L, LiM, ZhongG Y, et al. Circular trace transform and its PCA-based fusion features for image representation[J]. IET Image Processing, 2018, 12(10): 1797 - 1806. |
23 | AlbukhanajerW A, JinY C, BriffaJ A. Classifier ensembles for image identification using multi-objective Pareto features[J]. Neuro- computing, 2017, 238: 316 - 327. |
24 | OrdentlichO, ErezU. Performance analysis and optimal filter design for sigma-delta modulation via duality with DPCM[J]. IEEE Transactions on Information Theory, 2019, 65(2): 1153 - 1164. |
25 | YiD, LeiZ, LiaoS C, et al. Learning face representation from scratch[EB/OL]. , 2019-11-13. |
26 | Frías-VelázquezA, OrtizC, PižuricaA, PhilipsW, et al. Object identification by using orthonormal circus functions from the trace transform[A]. IEEE International Conference on Image Processing[C]. FL, USA: IEEE. 2012. 2153 - 2156. |
27 | LiuW Y, WenY D, YuZ D, et al. Sphereface: Deep hypersphere embedding for face recognition[A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. Hawaii, USA: IEEE Computer Society,2017.212 - 220. |
28 | SimonyanK, ZissermanA. Very Deep Convolutional Networks for Large-Scale Image Recognition[EB/OL]. ,2014 |
29 | HeK M, ZhangX Y, RenS Q, et al. Deep residual learning for image recognition[A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. Location: Las Vegas, NV, USA: IEEE Computer Society. 2016.770 - 778. |
30 | HuangG, LiuZ, van der MaatenL, et al. Densely connected convolutional networks[A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. Honolulu, HI, USA: IEEE Computer Society, 2017. 4700 - 4708. |
31 | HuJ, ShenL, SunG. Squeeze-and-excitation networks[A]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C]. Salt Lake City, UT, USA: IEEE Computer Society, 2018. 7132 - 7141. |
32 | ChenY P, LiJ N, XiaoH X, et al. Dual path networks[A]. Ulrike von Luxburg. Proceedings of the 31st International Conference on Neural Information Processing Systems[C]. New York: Curran Associates Inc, 2010. 4470 - 4478. |
[1] | 曲立国, 陈国豪, 胡俊, 陈鹏. 单次扫描连通域分析算法研究综述[J]. 电子学报, 2022, 50(6): 1521-1536. |
[2] | 冯西安, 寇思玮, 谭伟杰, 毕杨. 水声信号处理中的稀疏表示理论及应用[J]. 电子学报, 2021, 49(9): 1840-1851. |
[3] | 林金朝, 李必禄, 李国权, 黄正文, 庞宇. 基于EWT和结构特征提取的心电信号R波识别算法[J]. 电子学报, 2021, 49(6): 1217-1223. |
[4] | 徐元博, 江祥奎, 李龙. 高阶对称差分解析能量算子在微弱轴承故障信号提取中的应用[J]. 电子学报, 2021, 49(4): 750-759. |
[5] | 鲍中新, 文成林, 马雪. 一种基于数据变化率的预处理及主元分析故障诊断方法[J]. 电子学报, 2021, 49(11): 2234-2240. |
[6] | 罗会兰, 陈鸿坤. 基于深度学习的目标检测研究综述[J]. 电子学报, 2020, 48(6): 1230-1239. |
[7] | 黄鸿, 王丽华, 石光耀. 面向高光谱遥感影像分类的监督多流形鉴别嵌入方法[J]. 电子学报, 2020, 48(6): 1099-1107. |
[8] | 翟颖, 陈渤. 基于稳健变分自编码模型的雷达高分辨距离像目标识别算法[J]. 电子学报, 2020, 48(6): 1149-1155. |
[9] | 张丞, 何坚, 王伟东. 空间上下文与时序特征融合的交警指挥手势识别技术[J]. 电子学报, 2020, 48(5): 966-974. |
[10] | 郭小英, 李文书, 钱宇华, 白茹意, 贾春花. 可计算图像复杂度评价方法综述[J]. 电子学报, 2020, 48(4): 819-826. |
[11] | 刘颖, 刘红燕, 范九伦, 公衍超, 李莹华, 王富平, 卢津. 基于深度学习的小目标检测研究与应用综述[J]. 电子学报, 2020, 48(3): 590-601. |
[12] | 闻佳, 王宏君, 邓佳, 刘鹏飞. 基于深度学习的异常事件检测[J]. 电子学报, 2020, 48(2): 308-313. |
[13] | 李宝奇, 贺昱曜, 强伟, 何灵蛟. 基于并行附加特征提取网络的SSD地面小目标检测模型[J]. 电子学报, 2020, 48(1): 84-91. |
[14] | 彭丝雨, 周到, 张家琦, 王宇, 高军峰. 基于互信息的脑网络及测谎研究[J]. 电子学报, 2019, 47(7): 1551-1556. |
[15] | 钱晓亮, 白臻, 陈渊, 张鼎文, 史坤峰, 王芳, 吴青娥, 毋媛媛, 王慰. 协同视觉显著性检测方法综述[J]. 电子学报, 2019, 47(6): 1352-1365. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||