基于自发表情数据集预训练的多任务深度网络表情识别方法

Multi-task deep network expression recognition based on pre-training of Natural emotional data sets

摘要:

为了更好地将现有深度卷积神经网络应用于表情识别,提出将构建自然表情图像集预训练和多任务深度学习相结合的方法。首先,利用社交网络图像构建一个自发面部表情数据集,对现有深度卷积神经网络进行预训练；然后,以双层树分类器替换输出层的平面softmax分类器,构建深度多任务人脸表情识别模型。实验结果表明,本文提出的方法有效提高了人脸表情识别准确率。

Abstract:

In order to better apply the existing depth convolution neural network to facial expression recognition, a method combining pre-training of natural facial expression image set and multi-task depth learning is proposed. Firstly, a data set of spontaneous facial expressions is constructed from social network images, and the existing deep convolution neural network is pre-trained. Then, a deep multi-task facial expression model is constructed by replacing the planar softmax classifier of the output layer with a double-tree classifier. The experimental results show that the proposed method can effectively improve the accuracy of facial expression recognition.