摘要
针对目前人脸表情识别鲁棒性较差的问题,提出一种结合改进的深度信念网络DBNs(deep belief networks)和堆叠自编码器SAE(stacked Auto-encoder)的人脸表情识别方法。采用改进的深度信念网络作为提取人脸表情特征的检测器,堆叠自编码器作为识别表情的分类器,实现在人脸表情识别研究中深度信念网络与堆叠自编码器的结合。在JAFFE(Japanese female facial expression)数据库和CK+(extended Cohn-Kanade)数据库上的人脸表情的识别率有明显提高,该方法可以较准确地识别人脸表情。
Traditional facial expression recognition method can not perform robustly. To overcome this deep belief networks and stacked auto-encoder to facial expression recognition were proposed. Improved deep belief networkswere taken as the detector for extracting facial expression feature. Stacked auto-encoder was taken as classifier for identifying fa-cial expression. The combination of the deep belief networks and stacked auto-encoder was sion recognition. Experimentll results show that facill expression recognitions on Japanese female facill expression dataset andCohn-Kanade dataset increase significantly. The method can more effectively identify facial expression.
出处
《计算机工程与设计》
北大核心
2017年第6期1580-1584,1601,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61370082
61572142)
广东省省科技计划基金项目(2013B021600013)
关键词
人脸表情识别
深度信念网络
堆叠自编码器
检测器
分类器
鲁棒性
facial expression recognition
deep belief networks
stacked auto-encoder
detector
classifier
robustness