激光与光电子学进展, 2020, 57 (14): 141005, 网络出版: 2020-07-28  

结合局部二值模式和梯度特征的双通道表情识别

Double-Channel Facial Expression Recognition Based on Local Binary Pattern and Gradient Features
作者单位
中国民航大学电子信息与自动化学院,天津 300300
摘要
为了进一步提高表情识别准确率,提出一种结合局部二值模式(LBP)和梯度特征的双通道卷积神经网络表情识别算法。首先对采集得到的图像进行预处理,生成对应的梯度图像和LBP图像。针对单一特征对人脸信息表征不全面的问题,将特征提取网络分为两个通道,通道一输入梯度图像,提取人脸结构特征,从而更好地对人脸的全局信息进行描述,且对光照变化具有良好的鲁棒性;通道二输入LBP图像,提取人脸纹理特征以保留对五官边缘、亮点等微小特征的敏感性,两个特征相互补充,能够更加全面高效地对人脸特征进行表征,进而提高表情识别的准确率。最后通过加权特征融合网络对两种特征进行融合并利用Softmax对表情进行分类。在CK+、FER2013和Oulu-CASIA数据集上进行实验,分别取得了96.1%、75%和90.1%的平均识别率。结果表明,本文方法能够以较高的准确率识别6种基本面部表情,与单通道表情识别算法相比,取得了更高的识别准确率;相比于其他双通道卷积神经网络,能够以较简单的网络结构取得较好的识别效果。
Abstract
To further improve the accuracy of expression recognition, a double-channel convolutional neural network expression recognition algorithm combining local binary pattern (LBP) and gradient features is proposed. First, pre-processing approaches are implemented to generate corresponding gradient images and LBP images. Aiming at the problem of incomplete characterization of face information by a single feature, the feature extraction network is divided into two channels. One channel input gradient image to extract facial structural features so as to better describe the global information of the face and has good robustness to the illumination changes. Another channel uses the LBP images as input to extract the facial texture features to preserve the sensitivity to the edge of facial features, bright spots, and other micro features. Finally, the weighted feature fusion network is used to fuse the two features, and softmax is used to classify the expressions. Experiments implemented on CK+, FER2013, and Oulu-CASIA data sets have achieved average recognition rates of 96.1%, 75%, and 90.1%, respectively. The results show that the proposed algorithm can identify six basic facial expressions with higher accuracy than single-channel expression recognition method. Compared with other double-channel convolutional neural networks, it can achieve better recognition results with simpler network structure.

张红颖, 王汇三. 结合局部二值模式和梯度特征的双通道表情识别[J]. 激光与光电子学进展, 2020, 57(14): 141005. 张红颖, 王汇三. Double-Channel Facial Expression Recognition Based on Local Binary Pattern and Gradient Features[J]. Laser & Optoelectronics Progress, 2020, 57(14): 141005.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!