Research On Face Recognition Algorithm Based On EHMM-SVM

Now, universal application of the Internet and rapid development of science and technology, network security and information security issues become increasingly important. There are now a lot of means of identification based on biological characteristics, such as fingerprint, iris, face, etc. Face recognition is a common authentication method. Compared with fingerprints and iris, face recognition has some advantages, for example higher safety performance and recognition accuracy, easy to use, intuitive prominent, difficulty to counterfeit and fast identify. Face recognition is an important research area of pattern recognition. Select a good algorithm to improve the recognition rate of face recognition is an important and difficult problem.On the basis of previous studies and the knowledge of the integrated use of image processing and pattern recognition, EHMM-SVM face recognition algorithm is proposed in this paper. This paper mainly works in the following areas:? In this paper the image preprocessing common histogram equalization, image smoothing, image sharpening and geometric transformation method, combined with the ORL database image preprocessing experimental analysis. The results showed that: the use of histogram equalization and median filtering face gray image preprocessing could get better results. After histogram equalization face gradation processed image became clearer and bright. Median filter can eliminate noise and keep the details of the face grayscale image, so that the processed image is not showing a significant gray zone. Histogram equalization and median filtering method were used for image preprocessing.? In the facial feature extraction part, the paper often used for several facial feature extraction methods: K-L transform, wavelet transform, two-dimensional discrete cosine transform and singular value decomposition for a more detailed introduction, combined with the ORL face library on EHMM face model experiment analysis. The results showed that: the use of two-dimensional discrete cosine transform on the face image feature extraction could reduce the sensitivity to noise, light, changes in the expression and posture.? Reference Sammaria and Nefian human face gray image division model to study face recognition algorithm, combined with ORL and Yale face database experiment. The results of the analysis showed that: compared with the HMM recognition algorithm, EHMM face model to model in the horizontal direction, made better use of the linkages between local organ of the face, the face recognition rate had increased. However, when people face discrimination, EHMM model simply rely on maximum likelihood probability that is determined to face, there may be recognition errors. By SVM theory analysis, we found SVM is good at classification between categories; it is a good reflection of the different categories. Therefore, according to the characteristic of EHMM model and SVM, face recognition algorithm EHMM-SVM was proposed in this article.? ORL and Yale face image database were used to the experiment. Histogram equalization and median filtering method were used for image preprocessing; two-dimensional discrete cosine transform was used for human face gray image feature extracted to get the observation vector sequence. By double nested Viterbi algorithm derived for each face image corresponding to EHMM model output probability, the output of the input probability SVM trained in the classification and identification test to get the results of face recognition. Experimental results showed that based EHMM-SVM Face Recognition effectively improved the recognition rate. For face image deflection and light sensitive issue, which has not yet to be resolved but the recognition rate is improved fundamentally.