A hierarchical framework for image-based human age estimation by weighted and OHRanked Sparse Representation-based classification

Human age estimation based on face images can figure in a wide variety of real-world applications. In this paper, we propose a novel and efficient facial age estimation algorithm which decides human age in a hierarchical framework. Biologically, human lives can be roughly divided into two stages, the period from birth to adulthood and the period from adulthood to old age, which are quite different from each other in face growth and aging forms. Considering that craniofacial growth occurs mainly in the first stage while keeps basically stable in the second, based on the shape features, the coarse step of the framework determines which age stage the test sample belongs to using a quadratic function. On the other hand, since the face appearance of individuals in the same age group or even of the same age does have some similarities in common, accurate age estimation within the age stage is solved by Sparse Representation-based classification (SRC) in the fine step. However, SRC requires sufficient training samples in each class and in practice this assumption often does not hold, making the performance of age estimation limited. Accordingly, we take use of the idea of Ordinal Hyperplanes Ranker (OHRank) and weights of samples´ numbers in each class to solve the aforementioned problem, improving the age estimation results. Results of experiments conducted on the FG-NET Database demonstrate the effectiveness of our method.