结合LBP特征和深度学习的人脸表情识别

Facial Expression Recognition of LBP Features and Deep Learning

摘要:

对于人脸表情识别，传统方法是先提取图像特征，再使用机器学习方法进行识别，这种方法不但特征提取过程复杂且泛化能力也差。为了达到更好的人脸表情识别效果，文中提出一种结合特征提取和卷积神经网络的人脸表情识别方法。首先使用基于Haar-like特征的AdaBoost算法对于数据库原始图片进行人脸区域检测，然后提取人脸区域局部二值模式（Local Binary Patterns，LBP）特征图，将其尺寸归一化后输入到改进的LeNet-5神经网络模型中进行识别。在CK+和JAFFE数据集上采用10折交叉验证方法进行实验，分别为98.19%和96.35%的准确率。实验结果表明该方法与其他主流方法相比在人脸表情识别上有一定的先进性和有效性。

Abstract:

For facial expressions recognition, the traditional method is to execute feature extraction and recognize by machine learning. This method not only has complex feature extraction process but also poor generalization. In order to achieve better facial expression recognition, the paper proposes a facial expression recognition method combining feature extraction and convolutional neural network. Firstly, the AdaBoost algorithm based on Haar-like feature is used to detect the face region of the original image of the database, and then extract the local Binary Patterns (LBP) feature map of the face region, normalize the size and input it into the improved LeNet-5 network to recognize. The recognition rate is 98.19% and 96.35% respectively in the CK+ and JAFFE database with 10-fold Cross-validation method. The experimental results show that this method has certain advancement and effectiveness in facial expression recognition compared with other mainstream methods.