Local Gabor directional pattern for facial expression recognition

Humans display emotions through facial expressions. Detecting and recognizing these expressions comes naturally to us. But recognizing expressions is not straight forward for machines. In recent years a lot of literatures surged concerning automatic facial expression recognition. While some of these efforts have generated very good results, room for improvement still remains. There are several ways facial expression recognition can be performed. In this work we are proposing an appearance based method for extracting a new type of image descriptor based on Gabor wavelet which can be used for creating a feature vector of an image. To achieve this goal we applied Gabor filter over the image and recorded the edge response at each pixel. For each pixel we take several responses on several orientations. Finally we encode the responses and create a feature vector from the generated codes. We also showed experimental results on standard database for facial expression recognition using our proposed method.