부산대학교 도서관

堆积降噪自动编码机是一种典型的深度学习模型，它能够刻画数据丰富的内在信息，具有较强的特征学习能力。基于主成分分析（principal component analysis，PCA）技术和堆积降噪自动编码机（stacked denoising autoen-coders，SDAE）模型，提出一种新的表情识别算法PCA＋SDAE。该算法对人脸图片进行裁剪及归一化等预处理，采用主成分分析技术对人脸特征进行线性降维，再利用堆积降噪自动编码机逐层进行特征学习并同时实现对人脸表情数据的非线性降维，可以得到更好的、维度更低的表情特征，并据此进行表情分类。对PCA＋SDAE算法的仿真测试实验结果表明，其综合性能比其他的基于深度学习模型的表情识别方法更好，同时与传统的非深度学习表情识别方法相比，它具有更高的表情识别正确率。
A Stacked Denoising Auto-Encoders (SDAE)is a typical deep learning model.Because of its capability of dis-closing rich inherent information from data,and it has a strong ability of leaning features.Herein,a new algorithm princi-pal components analysis+stacked denoising auto-encoders (PCA+SDAE)for facial expression recognition is put forward on the bases of principal components analysis (PCA)technology and stacked denoising auto-encoders model.By the new algorithm PCA+SDAE,a facial image is firstly processed by cutting and normalization;then the linear dimensionality of its expression features is reduced by PCA;lastly,a greed layer-wise feature learning is conducted by a SDAE,and the non-linear dimensionality of its expression features is simultaneously reduced.Consequently,facial expression can be recognized based on the more powerful and lower dimensional facial features can be obtained.The results of simulation test experiments on algorithm PCA+SDAE show that the proposed method has better overall performance than some other expression recog-nition methods based on deep learning models;and it can also get higher expression recognition accuracy than traditional non-deep learning based expression recognition methods.