Multi-ethnic facial features extraction based on axiomatic fuzzy set theory

This paper proposes a new semantic concept extraction method to choose the salient features for representing multi-ethnic face characteristics based on axiomatic fuzzy set (AFS) theory. It has two advantages, one is that it could well convert the facial features to semantic concepts by bridging the semantic gap between image features and interpretable concepts; the other is that it could be considered as a dimension reduction method to preserve salient features for describing ethnic groups. Firstly, We build facial features to describe face with the landmarks of facial components, such as eyes, mouth and face contour, etc. , and then transform these facial features into semantic concepts. Secondly, a new approach is proposed to obtain the complex semantic concept sets of each ethnic group through clustering simple semantic concept based on AFS framework, and construct an optimal criterion to obtain valid semantic concepts of each ethnic group. Thirdly, we select the typical facial features which are corresponding to the semantic concepts to represent the ethnical face characteristic. Finally, we conduct experiments on Chinese Ethnic Face Database (CEFD), FEI and CK+ database to verify the effectiveness of our method. The K-means and fuzzy c-means (FCM) are used to verify the performance for describing multi-ethnic facial characteristics with the salient facial features. Specially, the obtained results demonstrate the efficacy of our approach, as the semantic concepts generated by optimal model can have an excellent interpretability and comprehension for the facial features. In addition, there is a comparative analysis between our method and other feature selection methods.