Real-time human cross-race aging-related face appearance detection with deep convolution architecture

Human age estimation (AE) is an emerging research topic in computer vision and machine learning and has attracted increasing amount of research due its wide potential applications. In the process of human aging, facial appearances change from glabrous to crinkly similarly across all races, from European, Hispanic and African to Asian. To specially explore the relationships between aging and facial appearances across races, this paper is devoted to determining the correspondence between facial aging and facial appearances. Specifically, we first extract appearance vector features from facial images with their spatial structure preserved. Then, we propose to select the aging-related features shared by different races to explore their aging-related common facial regions, while removing redundant features. Thirdly, we improve the proposed model by incorporating potential cross-race relationships in an automated learning manner. Additionally, we extend our model with deep convolution architecture. Finally, we evaluate the proposed methodologies on a large face aging database with real-time efficiency.