多尺度注意力机制DenseNet网络的表情识别方法

Expression Recognition Method Using DenseNet Based on Multi-scale Attention Mechanism

面部表情识别是人机交互研究领域的核心之一,现有的基于传统手工特征的面部表情识别方法难以应用在复杂多变的应用场景中。基于此,提出一种多尺度注意力机制的密集连接网络(DenseNet)表情识别方法。该网络模型对DenseNet121网络层数进行了简化,并引入多尺度结构和通道注意力模块MECANet,使得网络提取的面部表情特征更具判别性,有利于后续网络的表情分类。网络模型采用随机梯度下降算法进行训练,在CK+和FER2013数据集上取得了较高的识别率,分别达到96.2%和85.5%,与DenseNet121网络相比提高了8.4%和8.6%。

Facial expression recognition is one of the core of human-computer interaction research field. The existing facial expression recognition methods based on traditional manual features are difficult to be applied in complex and changeable application scenes. Based on this, an expression recognition method of DenseNet network with multi-scale attention mechanism is proposed. The network model simplifies the layers of DenseNet121 network, inserts multi-scale structure and channel attention module MECANet, which makes the facial expression features extracted by the network more discriminative and conducive to the expression classification of subsequent networks. The network model is trained by random gradient descent algorithm. High recognition rates are achieved on CK + and FER2013 data sets, reaching 96.2% and 85.5% respectively, which are 8.4% and 8.6% higher than DenseNet121 network.