【摘要】 现有方法识别精度受到大量与表情识别无关特征的影响,提出一种利用改进型GAN网络的面部表情识别。采用非对称局部二值模式提取特征;设计特征分离模型的改进Exchange-GAN网络,通过部分特征交换和约束实现表情相关特征和表情无关特征的分离,经过GAN分析实现面部表情识别;改进判别器与特征提取器间的对抗训练和内容训练,提高特征提取能力和面部表情识别的准确率。在3种数据集上对所提方法进行实验论证,其结果表明,该方法能够实现快速收敛,以FER2013数据集为例,其识别准确率较其它对比方法,分别提高了5.85%、4.13%和3.68%,具有较高的鲁棒性。更多还原

【Abstract】 In view of the fact that the recognition accuracy of existing methods is affected by a large number of features unrelated to expression recognition,a facial expression recognition method based on improved GAN network was proposed.Asymmetric local binary pattern was used to extract features.The improved Exchange-GAN network of feature separation model was designed.The expression related features and expression independent features were separated by partial feature exchange and constraint,and facial expression recognition was realized by GAN analysis.The confrontation training and content training between discriminator and feature extractor were improved to improve the ability of feature extraction and the accuracy of facial expression recognition.Experimental results on three datasets show that the proposed method can achieve fast convergence,and the recognition accuracy of the proposed method is 5.85%,4.13% and 3.68% higher than that of other comparison methods,respectively,with high robustness.更多还原