On Improving the Generalization of Face Recognition in the Presence of Occlusions

Xiang Xu, Nikolaos Sarafianos, Ioannis A. Kakadiaris

Abstract: In this paper, we address a key limitation of existing 2D face recognition methods: robustness to occlusions caused by visual attributes. To accomplish this task, we systematically analyze the impact of facial attributes on the performance of a state-of-the-art face recognition method and through extensive experimentation, quantitatively analyze the performance degradation under different types of occlusion. Our proposed Occlusion-aware face REcOgnition (OREO) approach improves the generalization ability of the facial embedding generator by learning discriminative embeddings despite the presence of such occlusions. The contributions of our occlusion-aware approach are two-fold. First, an attention mechanism is proposed that extracts local identity-related features from the global feature representations. The local features are then aggregated with the global representations to form a single facial embedding. Second, a simple, yet effective, training strategy is introduced to balance the non-occluded and occluded facial images. Extensive experiments with comparisons to strong baselines demonstrate that OREO improves the generalization ability of face recognition under occlusions by 10.17% in a single-image-based setting and outperforms the baseline by approximately 2% in terms of rank-1 accuracy in an image-set-based scenario.