Age-uniform Feature Learning for Image-based Kinship Verification

Kinship verification based on face image is an important topic in computer vision and has many applications in practice, such as family pedigree organization, missing person search, etc. Although parent and children share certain similarities in facial images, it is still difficult to verify the kin between people with large age gap. Therefore, how to reduce the influence of age factors on verification is the key to improve the accuracy of kinship verification. To this end, we propose an Age-uniform Face Representation Learning Network (AFRLN) to verify kinship. It mainly consists of Age Uniform Network (AUN) and Verification Network (VFN). Specifically, the design of AUN utilizes the idea of generative adversarial network, which aims to transform parent and child's face images of different ages into images of the uniform age range. Then, the transformed facial images are fed into the verification network, and discriminative deep features of parent and child are obtained. Finally, the output features are fused and then kinship verification task is conducted. Our approach is tested on two publicly kinship datasets: KinFaceW-I and KinfaceW-II. Experimental results validate performance of our method.