人脸表情分析是智能视觉信号处理和人工智能的重要研究领域，然而其研究仍然存在许多问题和挑战, 如人脸表情识别中局部关键点定位、系统的全局学习、自发表情数据缺乏、表情识别鲁棒性等问题。针对这些问题，本项目从特征表示、局部区域选择和表情识别角度出发，结合集成学习、深度神经网络、稀疏学习、生成对抗神经网络等理论，深入研究基于多深度网络融合和生成对抗网络的人脸表情识别与合成方法，主要集中在：1）针对联合局部区域选择和表情识别的面部表情特征学习问题，研究基于联合优化的多深度网络融合模型和基于稀疏集成的多深度网络融合模型； 2）针对深度网络训练中增强面部关键区域特征学习问题，研究基于增强深度网络的人脸表情识别方法；3）针对表情数据不足及表情不具自发性，研究基于多区域联合的生成对抗神经网络模型。本项目在提高人脸表情识别性能的同时，能进一步促进多深度模型融合和生成对抗网络的理论研究和应用。

Facial expression analysis is one curial research for intelligent visual signal processing and artificial intelligence, whereas there are many problems and challenges for this research, such as the local key positioning, the global optimization of the whole system of facial expression recognition, the lack of spontaneous expression data, the robustness of facial expression recognition, et al. For improving these problems, this project mainly focuses on the study of facial expression recognition based on multiple deep networks fusion and generative adversarial network from views of feature representation, local area selection and expression recognition, meanwhile by incorporating ensemble learning, deep neural network, sparse learning, generative adversarial network, et al, which includes: 1) for the problems that feature learning of facial expression by uniting the selection of local regions and expression recognizer models, we study the fusion models of multiple deep networks based on joint optimization and sparse ensemble; 2) for the problems of boosting the features of local key regions in the process of training one single deep network model, we study the boosted deep neural network model for facial expression recognition; 3) for the lack of facial expression data and the problem that expression is not spontaneous, we study generative adversarial network models based on joint multiple local regions. In this project, the studies of multiple deep networks fusion models and generative adversarial network models will be promoted, while improving the performance of facial expression recognition.