面向人脸表情识别的迁移卷积神经网络研究

摘要

人脸表情识别是模式识别研究的一个重要领域,现实环境中人脸表情识别容易受到光照、姿态、个体表情差异等因素的影响,识别效果仍有待提高.为了取得更好的人脸表情识别效果,本文提出一种基于迁移卷积神经网络的人脸表情识别方法,本文在训练得到人脸识别网络模型的基础上,采用迁移学习方法将所得人脸识别模型迁移到人脸表情识别任务上,并提出Softmax-MSE损失函数和双激活层(Double Activate Layer,DAL)结构,以提高模型的识别能力.在FER2013数据库和SFEW2.0数据库上的实验表明,本文所提方法分别取得了61.59％和47.23％的主流识别效果.%Facial Expression Recognition (FER) has always been an important field in pattern recognition.Because facial expression recognition can be easily affected by light,attitude and individual expression differences,facial expression recognition in the wild still did not obtain considerable progress.To achieve better facial expression recognition performance,a method of transferring face recognition net into facial expression net was proposed based on fine-tuning face recognition net.Furthermore,Softmax-MSE loss function and Double Activate Layer (DAL) structure were proposed to improve the discriminative ability of the model.The experiments were performed on FER 2013 dataset and SFEW 2.0 dataset and obtained overall classification accuracy of 61.59％ and 47.23％ respectively,which has achieved state-of-the-art performance.