Learning Deep Hierarchical Features with Spatial Regularization for One-Class Facial Expression Recognition

Existing methods on facial expression recognition (FER) are mainly trained in the setting when multi-class data is available. However, to detect alien expressions that absent during training, this type of cannot work. To address problem, we develop a Hierarchical Spatial One Class Facial Expression Recognition Network (HS-OCFER) which can construct decision boundary given class (called normal class) by training only one-class data. Specifically, HS-OCFER consists three novel components. First, hierarchical bottleneck modules proposed enrich representation power model and extract detailed feature hierarchy from different levels. Second, multi-scale spatial regularization with geometric information employed guide extraction towards emotional representations prevent overfitting extraneous disturbing factors. Third, compact intra-class variation adopted separate classes space. Extensive evaluations 4 typical FER datasets both laboratory wild scenarios show our method consistently outperforms state-of-the-art One-Class Classification (OCC) approaches.