Optimal Gabor Feature for Facial Expression Recognition

In order to reduce the curse of dimensionality of Gabor features in facial expression recognition, a two-level feature selection algorithm is developed.Firstly, the original Gabor features are pre-optimized according to the augmented variance ratio to represent the distinguish ability of each feature.Then, the most informative Gabor features are obtained with AdaBoost feature selection algorithm from the pre-optimized subset.As a result, the dimension of features is effectively reduced.Experimental results prove the effectiveness of the proposed method, by achieving a balance between training time and recognition rate.