Research Of Expression Recognition Based On Ehmm

With the rapid development of computer technology, the ways of human-computer interaction are gradually changing; intelligent human-computer interaction technology which emphasizes "people-oriented", "natural harmony" has been concerned widely. Intelligent human-computer interaction requires that the computer not only can be able to hear, see, say, but also has the ability of understanding, imitation and expressing human feeling. At the same time, human-computer interaction technology should be people-oriented, easy to use, makes it possible to use facial expressions, voice, gestures, body potential, lip movement and other natural ways to interact with the computer. Facial expression recognition technology is an important branch of human-computer interaction technology; applications of Facial expression recognition will fundamentally change the status of human-computer interaction, increase human feel of immersion tremendously in human-computer interaction. The relevant researches of Facial Expression Recognition relate to pattern recognition, image processing, affective computing and other fields, and it is a challenging topic.In this paper, some issues of Facial Expression Recognition are studied in-depth, and I propose a novel Facial Expression Recognition method which combining Gabor wavelet feature and the Embedded Hidden Markov (EHMM). Specific tasks are as follows: image preprocessing, completing related work of human face image gray uniform, face detection, eye location, scale normalized; proposing a novel eye location algorithm method by using the combining of template matching method and gray projection method, which compensates for the deficiencies of both and improves the accuracy of human eye location; in this part we also propose a new face segmentation method to achieve the face-scale normalization of expression face. Expression feature extraction, based on good features of Gabor wavelet, such as it isn't sensitive to light and can tolerate a certain degree of image rotation and deformation, five scales and eight orientations Gabor features are used to represent facial expression information. Classifier recognition, in this paper, we choose EHMM classifier to complete expression recognition, EHMM contains a super-state set and the corresponding one-dimensional HMM, its two-tier structure can be a good model two-dimensional image data. In the expression EHMM structure, the super states are used to model the expression image along vertical direction while the inner states are used to model the expression image along horizontal direction. Based on the testing and comparative analysis of the JAFFE database, we can see that the proposed method achieved a higher recognition accuracy, which can effectively recognize the basic expressions.