Person-Specific Face Spoofing Detection Based on a Siamese Network

Face spoofing detection is an essential prerequisite for face recognition applications. Previous face spoof-ing detection methods usually trained a binary classifier to classify the input face as a spoof face or a real face before face recognition, and client identity information was not utilized. In this paper, we propose a person-specific face spoofing detection method to employ client identity information for face spoofing detection. In our method, face spoofing is detected after face recognition rather than before face recog-nition; that is, the input face is recognized first, and the client identity is used to assist face spoofing detection. We train a deep Siamese network with image pairs. Each image pair consists of two real face images or one real and one spoof face image. The face images in each pair come from the same client. The deep Siamese network is trained by joint Bayesian loss together with contrastive loss and softmax loss. In testing, an input face image is recognized first, then the real face image of the identified client is retrieved, and an image pair is formed by the test face image and the retrieved real face image. The image pair is classified by the trained Siamese network to determine whether the input test image is a real face or not. The experimental results demonstrate the effectiveness of our method.(c) 2022 Published by Elsevier Ltd.

Automated summary

This paper proposes a person-specific face spoofing detection method that utilizes client identity information for improved detection accuracy. The method detects face spoofing after face recognition, using the identified client's face to assist in the detection. Experimental results demonstrate the effectiveness of the proposed method.