Table 1.
Overview of the two main components of systems for action unit recognition
| Study | Features | Classifier |
|---|---|---|
| Lien et al. (1998) | Dense-flow tracking | Hidden Markov model |
| Cohn et al. (1999) | Tracked feature points | Quadratic discriminant classifier |
| Fasel and Luettin (2000) | Eigenfaces | Nearest neighbor classifier |
| Bartlett et al. (2005, 2006) | Gabor filters | Boosting + support vector machine |
| Chang et al. (2006) | Manifold learning | Bayesian |
| Whitehill and Omlin (2006) | Haar features | Boosting |
| Littlewort et al. (2006) | Gabor filters | Boosting + support vector machine |
| Lucey et al. (2007) | Active appearance model | Support vector machine |
| Valstar et al. (2004) | Motion history images | Nearest neighbor classifier |
| Pantic and Rothkrantz (2004) | Tracked feature points | Rule base |
| Pantic and Patras (2005) | Tracked feature points | Rule base |
| Valstar and Pantic (2006, 2007) | Tracked feature points | Boosting + support vector machine |
| Tong et al. (2007, 2010) | Gabor filters | Boosting + dynamic bayesian network |
| Susskind et al. (2008) | Normalized pixels | Deep belief network |
| Koelstra et al. (2010) | Free-form deformations | Boosting + hidden Markov model |