Research On An Improved Facial Expression Recognition Method

With the increase needs of intelligence and the development of science and technology, it is a trend to make the computer to understand human emotions and people’s communication way. Facial expression is an important non-verbal communication method, it contains a lot of emotional information and reflects the inner world of human beings. So, the facial expression recognition technology has became a hotspot of the information processing and the concern of academics. Now, widely used in the field of distance education, clinical medicine, safety driving assistance, human-computer interaction, intelligent robots and so on.During the facial expression recognition process, feature extraction and classification step are both very important, they determine the final result. This paper proposes an improved facial expression recognition method based on the research of these two steps, mainly including following aspects:Firstly, how to obtain the most effective face features according to the needs of expression recognition. This paper proposes a hierarchical feature extraction method based on local binary pattern (LBP). This method combines the idea of subdivision and block, divides image into sub-blocks two times to extract the global and local information. During the stage of local feature extraction, use strategy to give the blocks involve nose, eyes, mouth organ a higher weight. Through this, it highlights these organ’s importance, getting a better description of the facial expression.Secondly, how to choose a right classification method which should be suit for the hierarchical feature extraction method. Besides, the combination of the two methods should also increase the recognition rate. After comparison and consideration, this paper chose the embedded hidden Markov model (EHMM) to be the classification method, it use the LBP histogram extracted from hierarchical feature extraction method to be the initial vector or observed sequence of the EHMM. Such a combination is effective to solve the problem that initial vector of EHMM model is difficult to determine. At the same time, fully take the advantage of hierarchical feature extraction method’s simple, rapid characteristics and EHMM’s high-precision, high-speed characteristics.Thirdly, how to determine hierarchical feature extraction method’s block way. In this paper, block way introduced is adaptable to an EHMM model. It divides the two-dimensional facial expression images into five parts and twenty-one sub-blocks from top to bottom. This is corresponding to EHMM’S five super-states and twenty-one sub-states. The relation of the two methods take full advantage of the most effective feature of two-dimensional facial expression and the two-dimensional nature of EHMM. Finally, make a great contribution to the overall recognition rate.Finally, this paper carry out the simulation experiment of expression recognition in the Matlab by using the JAFFE face database, This experiment use the way which combines the hierarchal feature extraction method and the Embedded Hidden Markov mode. Through the comparison of the associated method to prove the improvement and integration of LBP, EHMM is helpful for improving the recognition rate of facial expression.