Deep Facial Expression Recognition: A Survey

With the transition of facial expression recognition (FER) from laboratory-controlled to challenging in-the-wild conditions and recent success deep learning techniques in various fields, neural networks have increasingly been leveraged learn discriminative representations for automatic FER. Recent FER systems generally focus on two important issues: overfitting caused by a lack sufficient training data expression-unrelated variations, such as illumination, head pose, identity bias. In this survey, we provide comprehensive review FER, including datasets algorithms that insights into these intrinsic problems. First, introduce available are widely used literature accepted selection evaluation principles datasets. We then describe standard pipeline system with related background knowledge suggestions applicable implementations each stage. For state-of-the-art existing novel strategies designed based both static images dynamic image sequences discuss their advantages limitations. Competitive performances experimental comparisons benchmarks also summarized. extend our survey additional issues application scenarios. Finally, remaining challenges corresponding opportunities field well future directions design robust systems.