用于人脸表情识别的卷积神经网络研究

摘要

为了研究卷积神经网络在人脸表情识别中的应用，设计了一种10层的卷积神经网络模型识别人脸表情，最后一层用Softmax函数将表情的分类结果输出。首先，研究了卷积神经网络的卷积和池化算法并设计了模型的结构。其次，为了更形象地展示卷积层提取的特征，把提取的特征做了可视化处理并以特征图的形式展示。本文的卷积神经网络模型在Fer-2013数据集上进行了实验，实验结果展示了识别率的优越性。为了验证模型识别的泛化能力，最后自制了一个自然状态下的人脸表情数据集，并对人脸图片做了裁剪，灰度化以及像素调整等一系列的预处理。用本文模型识别该数据集中的人脸表情图片，识别的准确率达85.1010%。 In order to study the application of CNN in the field of facial expression recognition, the 10-layer CNN model is designed. The last layer of said model employs Softmax function to output the expression classification results. Firstly, this study concentrates on the convolution and pooling algorithm as well as the design structure of the model. In addition, the study visualized the extracted features and displayed them in the form of feature maps to show the features extracted by every convolutional layer. The study conducted experiments on the Fer-2013 dataset, and the result demonstrated the efficacy of the model. It is known that the Fer-2013 dataset contains data collected in an experimental environment. Therefore, to prove the effectiveness of the model, the study created a facial expression dataset by collecting facial expression images in a natural, spontaneous setting. The trained model, which was previously applied to the Fer-2013 dataset, was tested out on the new dataset. The experiment yielded promising results, one of which in the form of a recognition accu-racy rate as high as 85.1010%.