Research On Face Recognition Method For Multi - Expression

Face images are a type of special meaning images that can be obtained information by classification. So far, a face recognition is a very challenging task, it is an important research issue in computer vision and computer graphics science field. In the real life, people’s emotions can lead to changes in facial expression, which effect the face recognition efficiency. At present, multi-expression face images are widely used in animation and video production, identification, etc. In the future, there will be more extensive and broad application prospects. In these cases, it becomes critical to eliminate the effect of changes in facial expression during the face recognition process.In order to eliminate the influence of expression, we can do from the extraction of feature mainly. First, in the chaper 3, we use the face recognition method based on wavelet transform and principal component analysis, which extracted low frequency component information as input images. To some extent, this method can limited face expression, thus the recognition rate is increased. Then, in the forth chapter we the face recognition method of hidden Markov model, which is divided the human face into five parts:the forehead, eyes, nose, mouth and chin. In the feature extraction stage,it assignes different weights to the various parts, the effect face expression is decreased, and the recognition rate of face images is increased. Last, the wavelet transform and HMM are combined into a new approach, which can eliminate the influence of expression significantly, to achieve a high robustness, high efficiency multi-expression recognition.In this paper we use the Japanese ATR’s JAFFE face databases as the experimental samples, we compare the experiment’s results of the combined method of the wavelet transform and HMM with the combined of the wavelet transform and PCA, HMM recognition method, the experimental results show that:the face images use the wavelet transform to extract the low-frequency, then the PCA classification can get higher recognition rate than these method of based on the PCA,2DPCA and Module PCA; the combined method of the wavelet transform and HMM can has better recognition efficiency than the wavelet transform method, the HMM method. Experimental results proved that the combined method of the wavelet transform and HMM can achieve multi-expression recognition more better.