Facial Expression Recognition Algorithm Based On Static Facial Image

In recent years, with the rapid development of artificial intelligence technology, the demand for human-computer interactionwhich can be similar to the way of people communicate is increasingly strong. As a way of biometric identification, facial expression recognition is an indispensable part of human-computer interaction. Although this technology has made a lot of progress. However, many problems still exist, such as real –time is bad, correct classification rate is low, to achieve large-scale application needs further study. This paper, aiming at the research of facial expression recognition, on the basis of various algorithms, has been explored new and improved algorithm, which effectively increases the expression recognition rate and speed of execution. In this dissertation, the main research contents includes the following three aspects:The study based on rough set improved manifold constrained LLE algorithm. If we can find the control parameter expressions from "Face manifold", then we can greatly reduce the dimension of data preprocessing. In this paper,we studyed an algorithm which can be called Rough set Constrained Locally-linear Embedding, experiments shows that such constraints can effectively be further reduced dimensionality in order to avoid the LLE algorithm’s defec whicht is still need data redundancy. In the experimental of JAFFE expression database shows that the improved algorithm of this study, the facial expression of the seven categories, the average accuracy rate of 87.38%, this could be 5.95% higher than the result of Sander Koelstra.The study of the Non Subsampled Contourlet transform applied in facial expression recognition and design the corresponding application framework. First of all, the facial expression images that best characterize the image expression information is divided into two parts of the eyes and mouth, then use Non Subsampled Contourlet transform to extract feature,from the local image segmentation and finally the use of Extreme Learning Machine to classify the limit, with the BP neural network control experiment. The results showed that the average accuracy rate of expression classification is 86.57%, this could be 5.95% higher than the result of Sander Koelstra. BP neural network classification. In the method of execution speed, Extreme Learning Machine is 11.09 times than the BP neural network, whichi indicating the efficiency and feasibility of the experimental program.The study of the facial expression recognition algorithm based on the Visual word bag model. Different from the traditional way which is use the SIFT features extracted from the whole image as a visual dictionary, the innovation of this algorithm is based on the concept of the region of interest, using only information that best characterize the expression of the eyes and mouth to generate visual dictionary, that can greatly increase the expression classification accuracy. Finally, using Liblinear to classify expression images. Experimental results show that the average rate of the algorithm based on JAFFE expression database is 94.28%, 5.33% higher than the support vector machine, and the average time to run once only need 0.11 s, shortened 68 times than non-linear support vector machine; the average accuracy rate on CK expression database is 93.36%, 2.93% higher than the non-linear support vector machine, and the average running speed increases 14.25 times, that prove the generalization and efficiency of the algorithm.