Facial expression, which is a fundamental mode of transporting human's emotions, plays a significant role in our daily communication. Facial expression recognition is a complex and interesting problem, and finds its applications in driver safety, health-care, humancomputer interaction etc. Due to its wide range of applications, facial expression recognition has received substantial attention among the researchers in the area of computer vision [1-3]. Although a number of novel methodologies have been proposed in recent years, recognizing facial expression with high accuracy and speed remains challenging due to the complexity and variability of facial expressions. For facial expression recognition problems, the general recognition method appeared in previous work can be divided into two major steps, face representation and classifier construction. In the first step, features related to facial expression are extracted from images. Some of the features are hand-designed [4-6], whereas others are learnt from training images [7-9]. Then, the dimensionality of the features is reduced to facilitate an efficient classification and enhance the generalization capability. The universal expressions which are mentioned in the papers are usually anger, disgust, fear, joy (or happiness), sadness and surprise [1, 10], whereas some researchers add neutral as the seventh