Research On Face Recognition Based On2DPCA And FLDA

With the development of information science and technology, human biometric technologies have developed rapidly in the fields of security, military, civil and economic, such as fingerprint recognition, iris recognition, gene recognition, retina recognition. Compared with the above identification techniques, face recognition technology has many advantages, for instance, it’s very convenient, it hasn’t too much involvement, it hasn’t interference for the user, it’s not easy to be imitated, therefore, the research of face recognition which based on2DPCA and FLDA has very important significance.This paper build a static face recognition system, this paper do some deep research in image preprocessing, feature extraction, feature classification, and other related issues.First of all, this paper use some face image preprocessing methods such as histogram equalization, image sharpening and smoothing.Then, principal component analysis (PCA), two-dimensional principal component analysis (2DPCA) and Fisher linear discriminant analysis (FLDA) are itroduced and researched,2DPCA methodt isn’t converted to one-dimensional matri, it could process the two-dimensional image directly, it’s training time is less than PCA, we select2DPCA, at the same time, FLDA method is focused on classification, so we use a new method of2DPCA combined with FLDA, the method not only has the advantages of2DPCA,small amount of calculation and good reconstruction effects,but also has the advantages of FLDA,can be divided easily and select feature easily.This paper use a combined Feature classifier, which combined the nearest neighbor classifier with support vector machine classifier to recognize, let the feature vectors go into the combined classifier, using the nearest neighbor classifier to recognize roughly, and then using support vector machine classifier recognize narrowly, using the weighted result as the final classification and recognition results. The combined classifier is more better and more accurate than the single classifier. Last, this paper do some experiments on the ORL face database, and analysize the experimental results. Experiments show that feature extraction method based on2DPCA and FLDA achieves good recognition rate, and it provide reference value for the study of face recognition.