SOM-Based Continuous Category Learning for Age Classification by Facial Images

Abstract

In machine learning for person identification from facial images, feature sets are assigned into discrete categories. The age classification task, however, cannot be solved in the same way because the age-related face features go through gradual, or rather continuous, changes over the ages. From the machine learning viewpoint, age groups in terms of face images do not have clear borders, and it can be said that they can form ’continuous category’. Therefore we think that we should add the continuousness of face features to the learning process for enhancing the performance of age estimation. In this paper, we propose a model of age classification using one dimensional Self-Organizing Map, which can train the classifiers without preparing complete age information and reinforce the classifier’s performance. We show the effectiveness of our model and compare the performance to that of using discrete category machine learning. In conclusions, we clarify our future directions.