Re-evaluation of Face Anti-spoofing Algorithm in Post COVID-19 Era Using Mask Based Occlusion Attack

TL;DR

This work re-evaluates face presentation attack detection (PAD) under mask-based occlusions in the post-COVID era using Replay-Attack 2 and OULU-NPU 4 datasets. It compares four baseline PAD methods—LBP-SVM, QM-SVM, Frame Difference-SVM, and Deep-PiXBiS—plus a novel hybrid CNN-LBP model, under synthetic 2D/3D masks and glasses occlusions. Results show that occlusions significantly degrade all approaches, with handcrafted-feature methods suffering substantial increases in false rejections and false acceptances, while CNN-based methods exhibit greater robustness albeit still vulnerable to certain occlusions like glasses. The findings underscore the need for occlusion-aware PAD training data and the development of occlusion-agnostic detectors to ensure robust face authentication in real-world, mask-wearing conditions.

Abstract

Face anti-spoofing algorithms play a pivotal role in the robust deployment of face recognition systems against presentation attacks. Conventionally, full facial images are required by such systems to correctly authenticate individuals, but the widespread requirement of masks due to the current COVID-19 pandemic has introduced new challenges for these biometric authentication systems. Hence, in this work, we investigate the performance of presentation attack detection (PAD) algorithms under synthetic facial occlusions using masks and glasses. We have used five variants of masks to cover the lower part of the face with varying coverage areas (low-coverage, medium-coverage, high-coverage, round coverage), and 3D cues. We have also used different variants of glasses that cover the upper part of the face. We systematically tested the performance of four PAD algorithms under these occlusion attacks using a benchmark dataset. We have specifically looked at four different baseline PAD algorithms that focus on, texture, image quality, frame difference/motion, and abstract features through a convolutional neural network (CNN). Additionally we have introduced a new hybrid model that uses CNN and local binary pattern textures. Our experiment shows that adding the occlusions significantly degrades the performance of all of the PAD algorithms. Our results show the vulnerability of face anti-spoofing algorithms with occlusions, which could be in the usage of such algorithms in the post-pandemic era.

Figures (6)

• Figure 1: Example of spoof attacks with mask occlusions. First row depicts presentation attack images and second row depicts occlusion attacks by different kinds of masks.
• Figure 2: Example of the 68 facial landmarks predicted by the "dlib" 46 library.
• Figure 3: Figure depicting various 2D mask occlusions applied to the Replay-Attack dataset 2 and OULU-NPU dataset 4. Moving from left to right, the first column depicts a 2D mask with low-coverage ($\approx 30\%$) facial coverage). The mask covers the lower part of the face, and the nose of a person is visible. The second column depicts a 2D mask with medium-coverage ($\approx 40\%-50\%$ facial coverage) with some portion of the nose occluded. The third column depicts a 2D mask with high-coverage ($\approx 50\%-70\%$ facial coverage). One can see that a complete nose is blocked in this attack. Finally, the fourth column represents a 2D mask with round coverage ($\approx 30\%-40\%$ facial coverage).
• Figure 4: Figure depicting examples of mask occlusions with 3D cues on Replay Attack 2 dataset and OULU-NPU 4 dataset.
• Figure 5: Figure depicting an example of the glasses occlusion attack on Replay Attack (a) 2 dataset and OULU-NPU (b) 4 dataset.