Semantic facial descriptor extraction via Axiomatic Fuzzy Set

Abstract

In this paper, a semantic facial descriptor extraction method is proposed based on AFS theory with an aim to bridge the semantic gap between the low-level image features and high-level concepts. We first utilize the facial landmark detector to extract facial components automatically, such as eyes or nose. Then we propose a clustering algorithm based on Axiomatic Fuzzy Set (AFS) learning theory and cluster the detected facial components based on these detected landmarks. Finally we extract semantic descriptions for these facial components via assigning each facial component with semantic labels. The efficacy of this framework is demonstrated on two face datasets of Multi-PIE and BU-4DFE databases. The experimental results illustrate that the semantic facial descriptors obtained by the proposed AFS clustering technique are much better than those obtained by the conventional clustering techniques such as k-means and fuzzy c-means (FCM) in terms of consistency and comprehension, and they are much closer to human perceptions.