[얼굴 인식/Face Recognition] 얼굴 인식 관련 참고 자료

내가 보려고 만든 얼굴 인식(Face Recognition) 관련 참고 자료

 

지극히 주관적으로 필요한 자료들을 모아놓은 글.

 

1. 얼굴 인식 학습에 자주 사용되는 데이터셋 

얼굴 인식 학습에 자주 사용되는 데이터셋

 

출처: https://arxiv.org/pdf/1804.06655

 

2. 얼굴 인식 TECHNICAL CHALLENGES(기술적 한계)

• Security issues.

Presentation attack [289], adversarial attack [280], [281], [290], template attack [291] and digital manipulation attack [292], [293] are developing to threaten the security of deep face recognition systems. 1) Presentation attack with 3D silicone mask, which exhibits skin-like appearance and facial motion, challenges current anti-sproofing methods [294]. 2) Although adversarial perturbation detection and mitigation methods are recently proposed [280][281], the root cause of adversarial vulnerability is unclear and thus new types of adversarial attacks are still upgraded continuously [295], [296]. 3) The stolen deep feature template can be used to recover its facial appearance, and how to generate cancelable template without loss of accuracy is another important issue. 4) Digital manipulation attack, made feasible by GANs, can generate entirely or partially modified photorealistic faces by expression swap, identity swap, attribute manipulation and entire face synthesis, which remains a main challenge for the security of deep FR.

 

• Privacy-preserving face recognition.

With the leakage of biological data, privacy concerns are raising nowadays. Facial images can predict not only demographic information such as gender, age, or race, but even the genetic information [297]. Recently, the pioneer works such as Semi-Adversarial Networks [298], [299], [285] have explored to generate a recognizable biometric templates that can hidden some of the private information presented in the facial images. Further research on the principles of visual cryptography, signal mixing and image perturbation to protect users’ privacy on stored face templates are essential for addressing public concern on privacy.

 

• Ubiquitous face recognition across applications and scenes.

Deep face recognition has been successfully applied on many user-cooperated applications, but the ubiquitous recognition applications in everywhere are still an ambitious goal. In practice, it is difficult to collect and label sufficient samples for innumerable scenes in real world. One promising solution is to first learn a general model and then transfer it to an application-specific scene. While deep domain adaptation [145] has recently been applied to reduce the algorithm bias on different scenes [148], different races [173], general solution to transfer face recognition is largely open.

 

출처: https://arxiv.org/pdf/1804.06655