基于约束性循环一致生成对抗网络的人脸表情识别方法

Facial expression recognition method based on constraint cycle-consistent generative adversarial networks

深度神经网络在表情分类识别中会由于样本数量不均衡,样本相似度高出现过拟合,导致识别结果不理想,而通过旋转变换、添加噪声等数据增强方法不能解决相似度较高的问题。为此,对循环一致生成对抗网络(Cycle-GAN)进行改进,通过引入类别约束性条件,构建一种基于约束性循环一致生成对抗网络(CCycle-GAN),实现了一对多的数据类别映射,减少了模型训练开销;同时,为了提高人脸表情识别效率,对Cycle-GAN判别器进行改进,用一个辅助表情分类器替换循环一致生成对抗网络的判别器,不仅可以判断输入图像的真假,而且可以对表情进行分类。在CK+和FER2013数据集上的实验结果表明,该方法可以有效地解决神经网络中的过拟合和样本不均问题,提高了表情样本生成质量,进而提高了人脸表情识别率。

Due to the similar features of samples and the imbalance of the sample category,over-fitting occurred in the expression classification and recognition of deep neural network,resulting in unsatisfactory experimental results. Data enhancement methods such as rotation transformation and noise addition cannot solve the problem of high similarity. To this end,this paper improves the cycleconsistent generative adversarial networks( Cycle-GAN) and constructs a constraint cycle-consistent generative adversarial Networks( CCycle-GAN) by introducing the class-constrained condition,which implements one-to-many data category mapping and reduces the costs of model training. At the same time,in order to improve the efficiency of facial expression recognition,the discriminator of CycleGAN is replaced by an auxiliary facial expression classifier. This improvement not only discriminates the authenticity of the input image,but also classifies the expression. The experimental results on the CK + and FER2013 datasets show that the proposed method can effectively solve the problem of over-fitting and sample imbalance,also improve the quality of generated images,thereby improve the recognition rate of facial expressions.