東北大学 大学院情報科学研究科 情報基礎科学専攻 計算機構論分野
(東北大学 工学部 電気情報物理工学科 情報工学コース)
青木・伊藤(康)研究室

IEEE Student Awards The Best Paper Prize

Face Anti-Spoofing with Intermediate Features from Vision Transformer

Mika Feng (Tohoku University) , Koichi Ito (Tohoku University) , Takafumi Aoki (Tohoku University) , Tetsushi Ohki (Shizuoka University) , Masakatsu Nishigaki (Shizuoka University)
2025年度電気関係学会東北支部連合大会, September 2025.
Abstract

Face recognition systems are designed to be robust against changes in head pose, illumination, and blurring during image capture. If a malicious person presents a face photo of the registered user, they may bypass the authentication process illegally. Such spoofing attacks need to be detected before face recognition. The major spoofing attacks are shown in Fig. 1, including the “print attack” in which a printed face image is presented and “display attack” in which a face video is displayed on a device. To detect such spoofing attacks, it is necessary to detect a local and/or global difference between the live face image and the spoofed face image such as texture and depth. Since local features can be extracted in the shallow layer of Vision Transformer (ViT) and global features can be extracted in the deep layer of ViT, we investigate spoofing attack detection that takes advantage of such characteristics of ViT. Then, we propose a spoofing attack detection method that utilizes the intermediate features of ViT and introduces two data augmentation methods. We demonstrate the effectiveness of the proposed method through experiments using the SiW dataset.

戻る