解纠缠表示学习在跨年龄人脸识别中的应用

跨年龄人脸识别因其在现实生活中的广泛应用而成为人脸识别领域的热门话题.针对跨年龄人脸识别精度较低的问题,引入解纠缠表示学习,提出了一个基于生成对抗网络的解纠缠表示学习(IPDRL)网络来实现人脸图像的识别.该网络由编码器、生成器和鉴别器构成.编码器在对特征中的年龄变化进行解纠缠的同时,对人脸图像的身份信息进行编码,提取只利于身份鉴别的特征,实现身份特征和年龄特征的解纠缠;生成器根据输入的年龄特征生成对应的身份保持的年龄图像;鉴别器通过对抗学习和多任务学习实现年龄和身份的类分布预测.通过将解纠缠表示学习、对抗学习和多任务学习相结合的方法,很好地保留了人脸图像的身份信息,并使跨年龄人脸图像识别的精度得到了提高.

Application of disentangled representation learning in cross-age face recognition

Cross-age face recognition has become a hot topic in the field of face recognition because of its wide application in real life. In order to improve the accuracy of cross-age face recognition, this paper proposed an identity preserving disentangled representation learning(IPDRL) network based on generative adversarial network. The network consisted of encoder, generator and discriminator. The encoder disentangled any age change in the feature, and encoded the identity information of the image at the same time, and extracted features that were only conducive to identity identification, so that it realized the disentanglement of identity feature and age feature. The generator generated the corresponding identitied preserving age image according to the input age features, and the discriminator realized the class distribution prediction of age and identity through adversarial learning and multi-task learning. The combination of disentangled representation learning, adversary learning and multi-task learning can preserve the identity information of face image well, which improves the accuracy of cross-age face recognition.