基于超分辨率重建的低分辨率表情识别的研究  

Research on Low-resolution Facial Expression Recognition Based on Super-resolution Reconstruction

作　　者：王珏[1] 潘沛生[1] WANG Jue;PAN Pei-sheng(School of Telecommunications and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《计算机技术与发展》2021年第7期47-51,共5页Computer Technology and Development

基　　金：国家自然科学基金(61801240)。

摘　　要：人脸表情识别一直是人机交互和计算机视觉研究的热点问题,但受限于成像设备及环境的影响,获得的面部图像往往是低分辨率的。针对这个问题,提出一种基于深度学习的超分辨率重建的人脸表情识别系统,用于提高低分辨率面部图像表情识别的准确率。该系统是由两个深度神经网络组成,首个神经网络基于新的混合损失函数,利用简化的残差网络结构叠加构成残差块来学习面部图像的特征,并进行上采样操作,重建包含更多细节的高分辨率人脸表情图像;第二个神经网络通过使用小尺度卷积核提取重建后的高分辨率图像的人脸表情特征,之后使用softmax分类器,实现人脸表情分类。在公共数据集Cohn-Kanade Dataset (CK+)上对该系统进行验证和测试的结果表明,该系统有效地提高了不同尺寸的低分辨率图像表情识别的准确率。Facial expression recognition has always been a hot issue in human-computer interaction and computer vision research,but the facial images obtained are often low-resolution due to the impact of imaging equipment and the environment.To solve this problem,a facial expression recognition system based on deep learning super-resolution reconstruction is proposed to improve the accuracy of facial expression recognition in low-resolution facial images.The system is composed of two deep neural networks.The first neural network uses a simplified residual network structure to form a residual block to learn the features of the facial image,and performs an up-sampling operation based on a new hybrid loss function to reconstruct more detailed high-resolution facial expression images.The second one uses a small-scale convolution kernel to extract the facial expression features of the reconstructed high-resolution image,and then uses the softmax classifier to realize facial expression classification.The proposed system is verified and tested on the public data set Cohn-Kanade Dataset(CK+),which shows that this system improves the accuracy of the recognition of low-resolution expressions of different sizes effectively.

关 键 词：人脸表情识别 超分辨率重建 混合损失函数 低分辨率图像 小尺度卷积核 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]