人臉辨識系統的目的是自動化地將測試影像與資料庫的人臉影像做比對。然而，測試照片與資料庫照片是於不同時間點拍攝，比對的正確率會受限於攝影機拍攝角度(鏡頭縮放)或物件遮蔽等因素而降低。舉例來說，警察在追查從監視系統中取得的一些嫌疑犯照片，但這些嫌犯可能是有穿戴口罩或是穿戴墨鏡，這種大面積遮蔽導致警方難以與資料庫做比對找出兇手。本論文提出的方法是將大面積遮蔽的影像修復，修復影像群能幫助提升辨識率。我們所提出的自動化系統整合了人臉遮蔽恢復、人臉特徵點定位、人臉遮蔽區域偵測與辨識模組。

The purpose of the face recognition system is to automatically match the test image with the face image in the database. However, these two images are different are taken under different conditions, which decrease the recognition performance. For example, the suspect images captured by the surveillance system may wear scarf or sunglasses. This large-area missing makes it difficult for matching process. In this paper, we propose a occluded face recovery framework to improve the recognition rate. Our proposed framework integrates occluded face recovery, facial feature point prediction, occlusion area detection, and recognition modules.

目錄
圖目錄	V
表目錄	VI
第一章 緒論	1
1.1 研究目的與動機	1
1.2 相關研究	2
第二章 VAE RESBLOCK MODEL為主的人臉修復技術	4
2.1 IMAGE TO LANDMARK模型	5
2.2 IMAGE TO IMAGE模型	7
2.3 人臉修復系統流程	9
第三章 實驗結果	17
3.1 實驗設置	17
3.2 模型訓練之策略	17
3.3 遮蔽影像之人臉辨識率	18
第四章 結論	19
參考文獻	20
附錄:英文論文	22

圖目錄 
圖1.  VAE RESBLOCK MODEL 訓練時的模型	5
圖2. 整體系統流程	9
圖3. 修復影像結果示例	11
圖4. 修復影像藉由資料庫影像的輔助進行仿射變換，最後再依對應的權重加總合成	13
圖5. 最後真實遮蔽區域遮罩結果示例	14
圖6. 最後真實遮蔽區域遮罩結果示例	16

表目錄
表格1. 遮蔽影像之人臉辨識率與BASELINE比較	18

[1] H. Yang, D. Huang, Y. Wang, and A. K. Jain, “Learning face age progression: A pyramid architecture of gans,” Proc. IEEE Conf. Comput. Vis. Pattern Recognit, Jun. 2018.
[2] A. S. Jackson, A. Bulat, V. Argyriou, and G. Tzimiropoulos, “Large pose 3D face reconstruction from a single image via direct volumetric cnn regression,” in ICCV, 2017.
[3] Xing Di, Vishwanath A Sindagi, and Vishal M Patel. 2018. GP-GAN: gender preserving GAN for synthesizing faces from landmarks. In ICPR.
[4] X. Di and V. M. Patel. Face synthesis from visual attributes via sketch using conditional vaes and gans. arXiv:1801.00077, 2017.
[5] Johnson, J., Alahi, A., Fei-Fei, L.: Perceptual losses for real-time style transfer and super-resolution. In: ECCV. (2016)
[6] Larsen, Anders Boesen Lindbo, Sønderby, Søren Kaae, Larochelle, Hugo, and Winther, Ole. Autoencoding beyond pixels using a learned similarity metric. In ICML, 2016.
[7] X. Mao, Q. Li, H. Xie, R. Y. Lau, Z. Wang, and S. P. Smolley, “Least Squares Generative Adversarial Networks,” arxiv:1611.04076, 2016.
[8] M. S. M. Sajjadi, B. Scholkopf, and M. Hirsch. Enhancenet: Single image super-resolution through automated texture synthesis. In The IEEE International Conference on Computer Vision (ICCV), pages 4501–4510. IEEE, 2017.
[9] Isola, P., Zhu, J.-Y., Zhou, T., and Efros, A. A. (2016). Image-to-image translation with conditional adversarial networks. arXiv preprint arXiv:1611.07004 .