Human Identi cation versus Expression Classi cation via Bagging on Facial Asymmetry

We demonstrate a dual usage of quanti ed facial asymmetry for (1) human identi cation under expression variations and (2) expression classi cation across di erent human subjects. Our experiments show the e ectiveness of using statistical bagging and feature subspace selection BEFORE applying classi ers such as Linear Discriminant Analysis. This preprocessing allows the same type but di erent dimensions of image features to be discriminative for two seemingly con icting classi cation goals. Statistically signi cant improvements are found when facial asymmetry features are combined into classical classi ers.