基于个性化的人脸老化合成研究

一般的人脸老化合成方法往往针对不同个体的人脸会合成出非常相似的老化人脸，同时造成一些不受年龄影响的人脸属性的丢失或者发生变化，最终导致人脸老化目前还无法用来解决与年龄相关的人脸分析问题。鉴于此，本项目拟开展以下几个方面的研究：(1) 针对不同个体的老化合成人脸比较相似的问题，通过利用亲属人脸的老化特征来捕捉人脸的个性化老化模式，且进一步地利用相似人脸间的老化差异来加强这种个性化的老化模式;（2）针对部分不受年龄影响的脸部属性丢失或改变的问题，通过分离人脸属性中随年龄变化而变化的老化属性层与随年龄变化而不变的特定属性层，进而只对老化属性层进行老化重构;（3）针对跨年龄的人脸分析中老化差异的问题，通过利用人脸老化合成模型消除同一个人在不同年龄下的人脸面部属性老化差异。本项目研究不仅在人脸分析研究领域中具有重要的理论意义，而且能够推动人脸老化研究广泛的应用前景。

Existing face age synthesis methods often generate the similar aging faces for different individual faces, neglect some aging-invariant facial attributes in the age synthesis, and then cannot be collaboratively addressed the problems of the age-related face analysis well. Towards this end, the main works in this proposal include: (1) for the problem of the similar synthesized aging faces for different individual faces, we use the aging characteristics of kinship face to capture the personalized aging pattern of a certain face, and then utilize the aging difference of similar faces to enhance this personalized aging pattern; (2) for the missing or changing of some aging-invariant facial attributes, we divide a certain face into the aging-variant facial attribute layer and the aging-invariant facial attribute layer, and then reconstruct the aging-variant facial attribute layer at a future age; (3) for the problem of the aging difference in cross-age face analysis, we eliminate the aging difference of facial attributes between two faces by leveraging the face age synthesis model. Our research can not only bring in the important significance in the area of face analysis, but also improve the extensive application scenarios of age synthesis.