Learning Knowledge-based Prompts for Robust 3D Mask Presentation Attack Detection

TL;DR

This paper proposes a novel knowledge-based prompt learning framework to explore the strong generalization capability of vision-language models for 3D mask presentation attack detection, and incorporates entities and triples from knowledge graphs into the prompt learning process, generating fine-grained, task-specific explicit prompts.

Abstract

3D mask presentation attack detection is crucial for protecting face recognition systems against the rising threat of 3D mask attacks. While most existing methods utilize multimodal features or remote photoplethysmography (rPPG) signals to distinguish between real faces and 3D masks, they face significant challenges, such as the high costs associated with multimodal sensors and limited generalization ability. Detection-related text descriptions offer concise, universal information and are cost-effective to obtain. However, the potential of vision-language multimodal features for 3D mask presentation attack detection remains unexplored. In this paper, we propose a novel knowledge-based prompt learning framework to explore the strong generalization capability of vision-language models for 3D mask presentation attack detection. Specifically, our approach incorporates entities and triples from knowledge graphs into the prompt learning process, generating fine-grained, task-specific explicit prompts that effectively harness the knowledge embedded in pre-trained vision-language models. Furthermore, considering different input images may emphasize distinct knowledge graph elements, we introduce a visual-specific knowledge filter based on an attention mechanism to refine relevant elements according to the visual context. Additionally, we leverage causal graph theory insights into the prompt learning process to further enhance the generalization ability of our method. During training, a spurious correlation elimination paradigm is employed, which removes category-irrelevant local image patches using guidance from knowledge-based text features, fostering the learning of generalized causal prompts that align with category-relevant local patches. Experimental results demonstrate that the proposed method achieves state-of-the-art intra- and cross-scenario detection performance on benchmark datasets.

Figures (15)

• Figure 1: (a) The proposed knowledge-based prompt learning paradigm explicitly integrates expert knowledge in knowledge graphs and causal representation learning for prompt learning. (b) Common prompt learning only includes category names for business-related contexts.
• Figure 2: The proposed framework comprises two key components: a knowledge graph prompt module and a spurious correlation elimination module. The knowledge graph prompt module integrates entities and triples from the knowledge graph, along with learnable contexts, as fine-grained prompts to effectively adapt pre-trained vision-language models for 3D mask presentation attack detection. Meanwhile, the spurious correlation elimination module employs causal prompt learning to explicitly identify and remove category-irrelevant local image patches, thereby enhancing the model's generalization capability.
• Figure 3: Overview of the knowledge graph constructed for 3D mask presentation attack detection.
• Figure 4: Examples of fine-grained discriminative descriptions. Each quotation-marked phrase in the question illustrates an instance of a triple.
• Figure 5: Structure of the visual-specific knowledge filter.