Age classification system with ICA based local facial features

Abstract

This paper proposes an automatic age classification system that distinguishes kid faces from adult faces in real time. The system consists of three parts: face detection, face alignment and normalization, and age classification. We use standard face detection and face alignment method to generate face samples by automatically locating, cropping and aligning faces from images. We use ICA to extract statistically independent basis images which contain local facial components. Therefore, an image can be represented as a linear combination of those basis images. Then we choose a subset of basis images based on mutual information between them and class labels. Finally, we perform classification by SVM. Our experiment results show that our method provides better classification accuracy than conventional 1D or 2D projecting method. © 2009 Springer Berlin Heidelberg.