Research Of Facial Expression Recognition Algorithms

Facial Expression Recognition (FER) is an extremely challenging research project, it has caused a lot of attention because of its potential use value. A typical Facial Expression Recognition system usually contains the following three parts:face detection, facial expression feature extraction and classification of facial expression.In this paper, based on a survey of existing classic algorithms for facial expression recognition a detailed study about three parts of facial expression recognition is given.The first part is the face detection. In this part, firstly, color skin area is separated from multi-colored face image in YCbCr color space according to skin-color clustering, and then skin area image is got after binary image processing, in order to process holes left, morphological method is used. Finally, candidate face area needs to be checked, two methods is used in this step, one is known knowledge about face geometry, the other is locating eyes and lips in face area. At eye location stage, a new method that integral projection and the differential projection basing on known knowledge about face is proposed, and finally, simulation experiments show the effectiveness of this method.The second part is feature extraction of the facial expression. The original LBP-based method is sensitive to noise, therefore a new method is proposed to overcome the defect. Average value of center pixel and neighbor pixels are selected to be the threshold value.58 uniform patterns are used to describe facial expression, local facial expression feature and global facial expression feature are combined together, face area is divided into several regions, LBP histogram is computed over every sub-region, and they are concatenated into a single histogram. Since different regions of face area contains different facial expression feature, weights strategy is used to solve the problem, so a better facial expression feature representation is got.The third part is expression classification. SVM (Support Vector Machine) is a widely used method in pattern recognition field, and it is used as the classifier in this paper. RBF is chosen as the kernel function through experiments.301 facial expression images is used to train and test by SVM, and proves that improved LBP operator is better than the original LBP operator.