Abstract:

In 2017, artificial intelligence was officially upgraded to China’s national strategy. As an important research direction in the field of artificial intelligence, facial expression recognition has received extensive attention from domestic and foreign researchers. However, the traditional facial expression recognition technology cannot adapt the requirements of facial expression recognition in the natural environment. Therefore, non-frontal facial expression recognition methods have become the focus of realizing breakthroughs in the practical use of expression recognition technology. However, the existing research on non-frontal expression recognition faces many difficulties：the head deflection not only causes the distortion of the recognition image, but also partially occludes the face area, which seriously interferes with the extraction and recognition of expression features. In view of this, researchers combined deep learning technology with non-frontal facial expression recognition, relying on the depth information of non-frontal facial expression images to achieve the improvement of algorithm recognition capabilities. This survey introduces the structure of deep neural networks in detail, and compares the latest research methods of deep learning neural networks, meanwhile looks forward to future research and challenges.

Key words: non-frontal facial expression recognition, deep learning, neural network

摘要：

2017年人工智能正式升级为中国国家战略，作为人工智能领域中重要的研究方向，人脸表情识别受到了国内外研究者们的广泛关注。然而传统的人脸表情识别技术无法适应自然环境下的表情识别需求。因此非正面人脸表情识别方法成为实现表情识别技术实用化突破的重点。但是现有的非正面表情识别研究面临很多困难：头部偏转不仅造成了识别图像的扭曲，而且还遮挡了部分人脸区域，严重干扰了表情特征的提取与识别。有鉴于此，研究者们将深度学习技术与非正面表情识别相结合，依靠非正面表情图像的深度信息，实现算法识别能力的提升。综述详细介绍了深度神经网络的结构，对最新的深度学习神经网络研究方法进行分类对比，同时对未来的研究和挑战做了展望。

关键词: 非正面表情识别, 深度学习, 神经网络