Research On Facial Expression Recognition

Facial Expression Recognition (FER), which is the short title of the Automatic Facial Expression Recognition, refers to the process that people use computer technology to extract the feature information of Human Facial Expression, and class the information in accordance with understanding and ways of thinking of human, and analyze the emotion. FER is an important part of the Computer Vision Research.Feature Extraction is the core of Facial Expression Recognition. Automatic Facial Expression Recognition is difficult, because the face is a deformable body and difficult to build a precise mathematical model. The expression will change greatly when the place of organs in the face changes lightly. Therefore, the vital factor of deciding the recognition accuracy is how to select the features from human faces. Focusing on Feature Extraction, we analyze and research the Human Facial Expression Recognition deep in two ways of the Single-feature Extraction and the Multi-feature Fusion.Within-Class PCA is the process, in which each type of training samples are processed in PCA, the distances of the within-class samples are made much narrow while the distance of inter-class samples are made much large. In this paper, we use the within-class PCA method for Facial Expression Recognition and compare the approach with inter-class PCA.To Expression Recognition, each type of Facial Expression Feature has its own limitations. As a result, only extracting a kind of feature won't win well recognition rate. For this reason, the paper presents a new expression feature that is Angle Change Geometry Feature, and fuses the Gabor Feature and Angle Change Geometry Feature.Experiments have proved that the recognition accuracy of within-class PCA is higher than inter-class PCA, while the recognition accuracy of the fusion is higher than inter-class PCA and within-class PCA.