Discrimination of Spontaneous Micro-expression Types Based on Adversarial Decomposition Convolutional Network
Aiming at the problems of the difficulty in extracting spontaneous micro-expression components and the low classification recognition accuracy,the adversarial decomposition convolutional network is proposed to realize the extraction and classification of spontaneous micro-expression components through game and cooperation between networks.The images of neutral face are used as the real samples of the discriminant network,and images of spontaneous micro-expression are used as inputting samples of the decomposition network.According to the adversary between networks,output images containing only spontaneous micro-expression components can be obtained.Then,transfer learning and classification of spontaneous micro-expression components are realized through the transfer network.The cross database experimental results of spontaneous micro-expression show that the classification accuracy is improved and it has the effect of overcoming the differences of race and skin color in different databases.
国家科技期刊平台
NSTL
万方数据
