Abstract
The appearance of human changes with age significantly. It is still a big challenge to estimate age from the appearance of facial images. In this work, we focus on the application of age estimation from face images. We try to improve the accuracy of age estimation from two ways. First, we propose facial feature projection from uncertain label. By considering the continuity of age labels in facial feature projection step, we extract better features to represent age characteristics. In addition, we propose Bayesian age estimation framework by combining probabilistic model and age regression. We can incorporate any extra information into the unified age estimation framework easily. From the experimental results, the proposed Bayesian age estimation using facial feature projection from uncertain label can perform better than other existing age estimation methods.