Age estimation using local binary pattern kernel density estimate

Abstract

We propose a novel kernel method for constructing local binary pattern statistics for facial representation in human age estimation. For age estimation, we make use of the de facto support vector regression technique. The main contributions of our work include (i) evaluation of a pose correction method based on simple image flipping and (ii) a comparison of two local binary pattern based facial representations, namely a spatially enhanced histogram and a novel kernel density estimate. Our single- and cross-database experiments indicate that the kernel density estimate based representation yields better estimation accuracy than the corresponding histogram one, which we regard as a very interesting finding. In overall, the constructed age estimation system provides comparable performance against the state-of-the-art methods. We are using a well-defined evaluation protocol allowing a fair comparison of our results. © 2013 Springer-Verlag.