基於單邊元三元組損失函數的多任務架構應用於泛化人臉防偽辨識

莊筑鈞

18.207.132.226，Hello！

Close
Search Quick Search Advanced Search Publication Search	Personal Service Member Login Searching History Use Bonus Card SYMSKAN
Browse Browse by University of Graduate Thesis Browse by Pinyin of Graduate Thesis Browse by Conference Proceedings Topic Browse by Conference Proceedings Publisher Browse by Pinyin of Conference Proceedings Browse by eBook Category Browse by eBook Topic Browse by eBook Publisher
Customer Services Authorize Airiti Recommend a Journal

Search Symbol (Half-width)	Description of Search Symbols
Space	"AND" indicates the intertwining of key terms used in a search
Double Quotation Marks ("") ( " " )	Double quotation marks indicate the beginning and end of a phrase, and the search will only include terms that appear in the same order of those within the quotations. Example: "image process" ： " image process "
?	Indicates a variable letter. Entering two ? will indicate two variable letters, and so on. Example: "Appl?", search results will yield apple, apply… e , appl y … ( (often used to English word searches) )
*	Indicates an unlimited number of variable letters to follow, from 1~n. Example: Enter "appl*", search results will yield apple, apples, apply, applied, application…(often used in English word searches) e , appl es , appl y , appl ied , appl ication … ( (often used to English word searches) )
AND、OR、NOT	Boolean logic combinations of key words is a skill used to expand or refine search parameters. (1) AND (1) AND: Refines search parameters (2) OR (2) OR: Expands search parameters (3) NOT: Excludes irrelevant parameters

DOI stands for Digital Object Identifier （ D igital O bject I dentifier ） ,
and is the unique identifier for objects on the internet. It can be used to create persistent link and to cite articles.

Using DOI as a persistent link

To create a persistent link, add「http://dx.doi.org/」「 http://dx.doi.org/ 」 before a DOI.
For instance, if the DOI of an article is 10.5297/ser.1201.002 , you can link persistently to the article by entering the following link in your browser: http://dx.doi.org/ 10.5297/ser.1201.002 。
The DOI link will always direct you to the most updated article page no matter how the publisher changes the document's position, avoiding errors when engaging in important research.

Cite a document with DOI

When citing references, you should also cite the DOI if the article has one. If your citation guideline does not include DOIs, you may cite the DOI link.

DOIs allow accurate citations, improve academic contents connections, and allow users to gain better experience across different platforms. Currently, there are more than 70 million DOIs registered for academic contents. If you want to understand more about DOI, please visit airiti DOI Registration （ doi.airiti.com ）。

Data Source

Basic & Applied Sciences > Information Science
College of Electrical Engineering & Computer Science > Graduate Institute of Information Systems and Applications

Bibliography Management Tool

Related Links

Report an Issue

Purchase Single Articles

Download PDF

基於單邊元三元組損失函數的多任務架構應用於泛化人臉防偽辨識

Multi-Task Framework for Generalized Face Anti-Spoofing with One-Side Meta Triplet Loss

莊筑鈞 , Masters　　Advisor：賴尚宏　　

英文

電腦視覺；深度學習；人臉防偽辨識；多任務；元學習；域名泛化； Computer vision ； Deep Learning ； Face anti-spoofing ； Multi-task ； Meta learning ； Domain generalization

Abstract │ Reference (115) │ Altmetrics

Abstract 〈TOP〉

Parallel Abstract 〈TOP〉

Due to increasing variations of presentation attacks, model generalization becomes
an essential challenge for face anti-spoofing. Many previous works could not perform
well in generalization. This paper improves the generalization ability of face
anti-spoofing with two aspects. First, employing the face parsing information encourages
the network to focus on face regions and realizes distributions between
different face parts. Second, one-side triplet loss is adopted into the network to cooperate
with the meta learning process. This paper proposes a novel multi-task face
anti-spoofing framework that contains three tasks: depth estimation, face parsing,
and live/spoof classification. With the pixel-wise supervision from the face parsing
and depth estimation tasks, the regularized features can better distinguish spoof
faces. While simulating domain shift with meta learning techniques, the proposed
one-side triplet loss can further improve the generalization capability by a two-stage
margin setting. Our framework consists of a feature extractor, a depth estimator, a
U-net based face parsing module, and a meta learner for conducting meta learning
and classification. The proposed U-net based face parsing module contains a U-net
for predicting semantic face image and an attention-based skip connection for aggregating
the semantic information of different channels. Extensive experiments on
four public datasets demonstrate that the proposed framework and training strategies
are more effective than previous works for model generalization to unseen domains.
The AUCs are improved by over 6% compared to the baseline for some experiments
on domain generalization benchmark for face anti-spoofing, and the HTER is also
significantly improved over the previous methods.

Reference ( 115 ) 〈TOP〉

[1] Atoum, Y., Liu, Y., Jourabloo, A., and Liu, X. Face anti-spoofing using patch
and depth-based cnns. In In Proceeding of International Joint Conference on
Biometrics (2017).
[2] Balaji, Y., Sankaranarayanan, S., and Chellappa, R. Metareg: Towards domain
generalization using meta-regularization. In Advances in Neural Information

Altmetrics

〈TOP〉

E-mail ：

When an article is available to download, a notice will be sent to your mailbox address.

E-mail ：