深度残差网络和LSTM结合的图像序列表情识别

为了改善图像表情和图像序列表情识别效果，针对传统表情识别特征提取复杂和效果不理想问题，提出了一种深度残差网络和局部二值模式(local binary patterns,LBP)相结合的特征提取方法，利用深度残差网络提取数据集的空域特征，长短期记忆网络(long short-term memory,LSTM)处理时域特征，实现空域与时域特征的结合。研究了不同层数的残差网络、不同形式的LBP算子以及其他网络结构对人脸表情识别的影响，对比了支持向量机和随机森林实现的序列表情识别算法。在Cohn-Kanade数据集和AFEW6.0数据集上进行了验证，实验结果表明，算法在验证集上的准确率分别为73.1%和58.4%，相比其他算法有一定程度的提升。

Expression recognition of image sequence based on deep residual network and LSTM

In order to improve the effect of image expression and image sequence expression recognition, in view of the complexity and low efficiency of traditional expression recognition feature extraction, a feature extraction method combining deep residual network and local binary patterns (LBP) is proposed. The deep residual network extracts the spatial features, and the long short-term memory (LSTM) network processes the temporal features, which combines spatial and temporal features. The effects of different layers of residual networks, different forms of LBP operators and other network structures on facial expression recognition are studied. The support vector machine and random forest implementation are compared. The algorithm is verified on the Cohn-Kanade data set and the AFEW6.0 data set. The experimental results show that the accuracy of our algorithm on the Cohn-Kanade and AFEW6.0 verification sets are 73.1% and 58.4%, respectively. Compared with other algorithms, this algorithm has a certain degree of improvement.