FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models

FaceShield：The face in the image is a spoof, as it is cleverly concealed by a paper with printed eyes, which is a common technique used to create a fake appearance. Users：Is the face in the image real or spoof ? Spoof FaceShield：According to the picture, from the "Face Facial Lighting" the lighting seems uneven and unnatural, suggesting artificial illumination. The "Global Shape Consistency of Facial Features" appears distorted, lacking natural symmetry. The "Sense of Depth and Three-Dimensionality" is flat, missing the natural depth cues found in real faces. Lastly, the "Presence of Phone Screen or Paper Edges" is evident, indicating that this face is displayed on an electronic device. Analyzing these attributes, we can conclude that the face in the picture is a spoof. Users：Why is the face in the image real or spoofed? Coarse-grained classification FaceShield：The face in the image is a Print Attack. This is evident from the appearance of the paper edges in the picture, indicating that a printed face is used to deceive the face detection system. Users：What Attack Type of the face in the image ? Print Attack Attack Type Fine-grained classification Spoof Reasoning FaceShield：The location of the spoof area is [ 99.0 295.0 341.0 485.0 ], which is the area where the person is holding a card with a printed lower half of their face to cover the real lower half of their face. Users：Can you locate the spoof area of the face in the image ? Localization Figure 1: FaceShield Multi-task Response Demonstration. This figure shows the model's performance on four tasks: coarsegrained classification (real vs. spoofed faces), fine-grained classification (specific attack types like print attacks), reasoning (explaining spoofing using features such as lighting and symmetry), and localization (detecting spoofed regions). It highlights FaceShield's ability to handle diverse, complex questions accurately.