A novel hierarchical regression approach for human facial age estimation based on deep forest

Human facial age estimation has received continuous attention in security control, age-targeted advertising, biometrics, etc. In this paper, a novel deep forest approach (named fcXGBoost, i.e., multi-features fusion cascade XGBoost) is proposed for the regression task of human facial age estimation. The fcXGBoost approach consists of Cascade XGBoost with a cascade structure for representation learning, and multi-feature fusion for the enhancement of its representational learning ability with fusion of multiple features (including Active Appearance Models (AAM), Local Binary Patterns (LBP) and Gabor Wavelets (GW)). In addition, a novel hierarchical regression approach is proposed to further improve the estimation performance, which consists of a rough regression stage and a detailed regression stage. Furthermore, a flexible overlapping of age ranges in the detailed age estimation is proposed to eliminate the influence of error in rough regression stage. The proposed approach is evaluated on the benchmark and public facial image dataset of MORPH Album 2 by comparison to the state-of-the-art results.