A robust framework for multiview age estimation

Age estimation from facial images has promising applications in human-computer interaction, biometrics, visual surveillance, and electronic customer relationship management, etc. Most existing techniques and systems can only handle frontal or near frontal view age estimation due to the difficulties of 1) differentiating diverse variations from uncontrollable and personalized aging patterns on faces and 2) collecting a fairly large database covering the chronometrical image series for each individual in different views. In this paper, we propose a robust framework to deal with multiview age estimation problem. A large face database, with significant age, pose, gender, and identity variations, is exploited in the experiments. In our framework, the training set is partitioned into small groups, namely code groups, according to their multiple labels, e.g. pose and age. Based on certain set-set distance measure, a compact representation for each image is obtained by measuring the distance between the image and all the code groups, which can be followed by classification or regression algorithms. Extensive experiments and comparisons with traditional multiview models demonstrate the proposed framework with significant advantages of variation decomposable, classifier adaptable, and feature selectable and extendable.