摘要
基于从一张给定年龄的人脸输入图片准确预测出该人老龄化后的人脸图片,并且保持身份的目的,采用了基于条件对抗神经网络的无监督跨领域框架,然后将此框架应用到人脸身份保持的老龄化的任务中的方法。所采用的对抗神经网络通过预训练的人脸识别网络提取源图片的特征,然后将目标年龄信息附加在嵌入式特征空间里,并且送往生成器,随之施加了身份保持的约束。所提出的算法在生成CACD人脸数据集的老化人脸生成实验上产生了高质量,并且保持住身份信息的人脸图片,同时,在人脸跨年龄分类任务上取得了2.64%的识别率增益,进而验证了算法的高准确率和有效性。
In this paper, based on the purpose of accurately predicting the synthesized aging face images from an original face image and preserving the identity information, we adopt an unsupervised crossdomain image generation framework based on a conditional Generative Adversarial Network, and apply it to the identity-preserved face aging task. To generate target domain alike images, we extract features of source images through a pre-trained face recognition network, then target age information is appended to the embedded features and sent to a generator, thus, an identity-preserved term is imposed. The proposed algorithm generates faces with high quality and preserves the identity information well.Meanwhile, compared with the best deep learning result on the CACD test set, the accurate rate of our model has been increased by 2.64%, which verifies its high accuracy and effectiveness.
作者
汤旭
TANG Xu(Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 200050, China;School of Information Science and Technology, ShanghaiTech University, Shanghai 201210, China;University of Chinese Academy of Sciences, Beijing 100049, China)
出处
《电子设计工程》
2018年第7期174-178,184,共6页
Electronic Design Engineering
