Facial Expression Recognition Based on Gabor Histogram Feature and MVBoost

In this paper, the Gabor transform is combined with the hierarchical histogram to extract facial expressional features, which could hierarchically represent the change of texture in local area, and thus capture the intrinsic facial features involved in facial expression analysis. In addition, Gabor transform is relatively less sensitive to the change of lighting conditions and can thus tolerates certain rotation and deformation of images. All these properties make this representation scheme quite robust in various conditions and more powerful than traditional representation scheme using 1-D Gabor coefficients. With the histogram feature used, a weak classifier which uses vectors as input and multi-class real values as output is designed to classify facial expression. This weak classifier has been embedded into the multi-class real AdaBoost algorithm to meet the request of multi-class classification of facial expression. The resulted method (named MVBoost) directly assigns a multi-class label to every input image. Thus it needs not to train many two-class classifiers to meet the request of multi-class classification. The training process and classification process are both simplified. Experiments conducted on common database show the efficiency and effectiveness of the proposed technique. It is expected that this new technique will make contribution to fast and robust facial expression recognition.