A New Method for Age Estimation from Facial Images by Hierarchical Model
Li Zhang, Xianmei Wang, Yuyu Liang, Lun Xie · 2013
Different from the traditional age estimation systems, this paper introduces a novel method to estimate human age from facial images combined with non-fixed age group classification and Rank-based age value estimation. The basic idea is to use the continuous and ordinal information from human facial aging process. Firstly, pair-wise distance classifiers are employed to roughly estimate age within total database. Then a flexible and more precise age group is decided by extending the rough age value to a specific age scope. For the ranking framework of precious age estimation, by calculating the continental distances to select the most similar sample from the training database. The average label of the voted samples is used to predict age. In our proposed system, six-dimension shape features and different texture features are used to describe facial images. Tested on FG-NET database, our system achieves 4.89 evaluated by MAE (Mean Absolute Error).