Relative Ordering Learning Based on Weighted Sparse Representation for Age Estimation

摘要

Age estimation has received extensively interest in computer vision field due to its wide applications, i.e. human behavior analysis, video content recognition, etc. Hence, many classification and regression approaches were widely used to estimate human age, and obtained the basically acceptable results, but the relative relationships of age groups have not been exploited in previous methods. Recently, much work proposed to use various ranking-based learning techniques to exploit the relative ordering information. This paper presents a novel relative ordering learning algorithm based on weighted sparse representation to advance the accuracy of age estimation. The age estimation problem is decomposed into a series of relative age ordering sub-problems, and the relative ordering learning of each sub-problem is performed by using weighted sparse representation. Thus, the human age is inferred by aggregating all relative comparison results. The greatest advantage is that the weighted value can be embedded into the age estimator, and also can be dynamically updated in term of the distributions of the training data. The proposed approach is validated in extensively used aging database FG-NET by comparing it with some existing state-of-the-art methods. The experimental results show that the proposed approach outperforms other methods in all cases, and can effectively solve the imbalance and inconsistency of the training data.