Region-Aware Generalized Face Anti-Spoofing via Chebyshev Convolutional Graph Networks
Keywords: GNN, face spoofing, CNN
Abstract: Face Anti-Spoofing (FAS) is critical for safeguarding face recognition systems from spoofing attacks. However, current methods based on Convolutional Neural Networks (CNNs) and Vision Transformers face limitations in modeling the diverse, region-specific attack behaviors, leading to reduced generalization. This challenge arises due to two main factors: (1) attacks manifest differently across facial regions due to variations in color, texture, and material properties; and (2) the large data space hinders effective generalization.
To address these issues, we propose a novel approach utilizing Chebyshev Convolutional Graph Neural Networks (ChebConv GNNs) to model spatial information within a graph-based structure. ChebConv is particularly efficient in processing visual data from graphs derived from images. Our method processes regions around facial landmarks through the initial layers of a DenseNet to extract rich, local node features for each region. By assigning a node to each facial region, we construct a unified graph structure where the nodes correspond to the same regions across all faces. This enables the network to model local features and inter-region relationships effectively, reducing the data space and enhancing generalization.
To further improve generalization across unseen domains, we integrate a Domain-Adversarial Graph Network. Additionally, we introduce an auxiliary self-supervised task to encourage the learning of region-specific texture features. Experimental results demonstrate that our method significantly outperforms existing approaches in terms of both accuracy and generalization.
Primary Area: learning on graphs and other geometries & topologies
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7897
Loading