Based On The Overall Identification And Integration Of Local Recognition, Face Expression Recognition

Facial expression recognition is a growing scientific research hotspot in recent decades. It studies people's emotion by analyzing dependent facial expression with computer, and then to implement more natural and intelligent interaction between human and computer. It has potential applications in a great deal of domains, such as natural and harmonious human-computer interaction, artificial intelligence, machine vision, safe driving, security supervision in public place, distinguishing lie, computer games, and etc. So to analyze and recognize facial expression has significant meanings, and it is an important task that we have to solve in researching human-computer interaction.The paper has summarized research significance, applications and status of facial expression recognition. It reviews actual facial expression recognition methods and introduces face detection and image normalization. The paper has emphasized several following issues.1. It studies how recognition rate be affected by image compression and feature extraction dimensions of PCA. The experiment is implemented in JAFFE and CED-WYU(1.0) database. Several useful conclusions have been gained. First, higher recognition rate isn't always following bigger size of images compressed. Second, feature dimensions of PCA are concerned about classifiers. Third, the feature dimension of the highest recognition rate in different sizes of images is almost same. Four, highest mouth recognition rate is obtained in a low dimension. Those conclusions will benefit to next works in facial expression recognition.2. A new method adjusted by Hopfield network is proposed for expression recognition. First, a standard expression images should be selected for training of Hopfield network. Then, when Hopfield network has been trained, all expression images would be adjusted to close with similar standard images. This would increase distances between seven expression images. It would benefit for expression recognition. Experiment result shows this method can highly improve recognition rates.3. A method for face expression recognition is proposed by combining the global recognition and local recognition. The recognition is completed by combining in expression recognition results from the whole face and local regions such as mouth, eyes, and nose. This method has both advantages of global recognition and local recognition. Expression feature is based on PCA+LDA. Using LDC and KNNC separately to get global recognition and local recognition. And then the final expression recognition is obtained by combining the results of the whole face recognition and local recognition with Production combination rule and Dempster-Shafter combination rule separately. Experiment results show this method is validity.