Optoelectronic Technology, Volume. 43, Issue 1, 17(2023)
High Resolution Face Image Reconstruction Based on Deep Feature Embedding
Figures & Tables(13)
Fig. 7. Distribution of feature similarity when feature extractor for reconstruction and evaluation frameworks both are Inception
Fig. 8. Distribution of feature similarity when feature extractor for reconstruction and evaluation frameworks are Inception and Xception respectively
Table 1.
TAR when feature extractor for reconstruction and evaluation frameworks both are InceptionView tableView in ArticleTable 1.
TAR when feature extractor for reconstruction and evaluation frameworks both are Inception数据集 FAR/（%） 阈值 正常TAR/（%） Ⅰ型评估TAR/（%） Ⅱ型评估TAR/（%） LFW 0 0.531 6 79.77 100 64.00 0.1 0.438 2 98.12 100 96.04 1 0.400 0 99.42 100 99.17 10 0.352 2 99.86 100 99.72 ColorFeret 0 0.528 0 99.46 100 98.92 0.1 0.465 5 100 100 100 1 0.423 9 100 100 100 10 0.371 2 100 100 100
Table 2.
TAR when feature extractor for reconstruction and evaluation frameworks are Inception and Xception respectivelyView tableView in ArticleTable 2.
TAR when feature extractor for reconstruction and evaluation frameworks are Inception and Xception respectively数据集 FAR/（%） 阈值 正常TAR/（%） Ⅰ型评估TAR/（%） Ⅱ型评估TAR/（%） LFW 0 0.493 3 91.90 40 23.11 0.1 0.446 3 97.76 74 48.71 1 0.401 8 99.41 92 76.24 10 0.352 2 99.90 100 96.96 ColorFeret 0 0.507 8 99.64 54 43.55 0.1 0.481 9 99.82 70 58.06 1 0.430 0 100 90 86.56 10 0.373 8 100 100 99.46
Table 3.
TAR of different methods on LFW dataset at 0.1%FAR View tableView in ArticleTable 3.
TAR of different methods on LFW dataset at 0.1%FAR 方法 特征提取器 分辨率 Ⅰ型评估TAR/(%) Ⅱ型评估TAR/(%) 本文 ArcFace 1 024×1 024 100.00 96.04 Mapping ArcFace 1 024×1 024 1.42 0.46 NbNet FaceNet 160×160 95.20 53.91
Table 4. The pass rate of reconstructed images on living detect APIs
View tableView in ArticleTable 4. The pass rate of reconstructed images on living detect APIs
活体检测API 重建图像方法 通过图像数 未通过图像数 通过率/（%） API 1 NbNet 0 150 0 Mapping 129 21 86 本文 133 17 88.67 API 2 NbNet 70 80 46.67 Mapping 110 40 73.33 本文 112 38 74.67
Tools
Get Citation
Copy Citation Text
Zhihui MIAO, Yongai ZHANG, Zhixian LIN, Jianpu LIN. High Resolution Face Image Reconstruction Based on Deep Feature Embedding[J]. Optoelectronic Technology, 2023, 43(1): 17
Paper Information
Category: Research and Trial-manufacture
Received: Sep. 26, 2022
Accepted: --
Published Online: Apr. 14, 2023
The Author Email: LIN Jianpu (ljp@fzu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20