An advisable facial semantic characterization based on Axiomatic Fuzzy Set theory and information granules

Abstract With the ever-increasing popularity of social networks, a colossal amount of images are being uploaded to the digital world including human faces, and semantic facial descriptors would be very useful to describe some facial action units, which can be further used for affection analysis. In this paper, a novel facial semantic characterization method via Axiomatic Fuzzy Set (AFS) clustering scheme and information granules is proposed. Firstly, semantic descriptions for each facial component are extracted in the framework of the AFS theory. Then in order to further characterize the semantics obtained by the AFS clustering, the information granules are created around the constructed prototypes, that can specify the merits of the semantic facial descriptors. Multiple experiments on the Multi-PIE and BU-4DFE facial databases demonstrate that the proposed facial semantic characterization method not only can obtain more accurate and interpretable descriptions, but also the results are much closer to human perception.