一种深度自编码器面部表情识别新方法

为解决面部表情特征维度高的问题,该文提出了一种基于深度学习自编码器的表情识别新方法,该方法利用深度自编码器在多层隐层上进行特征选择,能够在较低维度上表示高维度的面部特征.首先采用定向梯度直方图从面部表情的选定区域提取特征,然后在多个层面上使用深度自编码器,得到最优编码特征,降低特征维度,最后使用支持向量机模型对降维特征进行分类.实验表明,与其他现有特征选择和降维技术相比,该文方法提取的特征优于其他特征,并能够有效实现面部表情识别.

A New Method for Facial Expression Recognition Based on Deep Autoencoders

To solve the problem of high facial expression feature dimensions, a new method for facial expression recognition based on deep autoencoders has been presented in this paper. In this method, autoencoders is used to perform feature selection on multi-layer hidden layers, which can be able to represent high-dimensional facial features in lower dimensions. Firstly, the histograms of oriented gradients is used to extract features from selected regions of facial expressions. And then deep autoencoders is used on multiple levels to obtain the best coding feature, reducing the feature dimension. Finally, the support vector machine model is used to classify the dimensionality reduction features. Experiments show that the features extracted from the deep autoencoder outperformed when compared to other feature selection and dimension reduction techniques, and can effectively realize facial expression recognition.