Research Of Face Recognition Based On2D-Neighborhood Preserving Embedding

Since the beginning of the21st century, with the rapid development of computerhardware and software, computer vision and pattern recognition become prevalent in imageprocessing technology quickly. More and more occasion’s need the admission personnelidentity, and the flow of people is relatively large, traditional authentication methods cannotmeet the demand in this era of increasingly information. So simple, prompt, efficient, and thecorrect face recognition technology came into being. And compared with other humanbiological characteristics like fingerprint recognition, iris recognition, voice recognition; facerecognition is even more direct, convenient and friendly. It is the characteristics of face moresuitable as the important distinguish information of people than any other biometric, becausethis view accord with the normal understanding of logic that one can tell the differencethrough eyes simply. Therefore, the emerging of face recognition technology quickly caused atremendous response in the computer industry, and it soon became a popular study of humanbiometric technology, the developing and application prospects of face recognition are bright.Both theoretical research and actual development has made a very good development, greatlyreduce the cost of human labor, and led the normalization and modernization of the associatedindustries.Over the past decades, the development of face recognition technology never stop, manyresearchers have proposed different methods to be applied to the face recognition technologyin the past years, and have achieved good effects, but there is still some problem notcompletely solved, for example, large amount of calculation, low recognition rate in someface database on some of the issues. There are many reasons for these problems, For example,the face image illumination, rotation, shelter, and other objective factors, those are the issuesour face recognition technology should consider and resolve.This paper presents a face recognition method based on two-dimensional nearestneighborhood preserving embedding. Compared with principal component analysis (PCA),the nearest neighborhood preserving embedding algorithm is a subspace learning methodbased on manifold, it is mainly consider the local nearest neighbor structure of database, thePCA main maintained the global data Europe Euclidean distance. The nearest neighborpreserving embedding considers local neighbor information of the data, and characteristicsneighbor of these data are maintained after projection. But the local features of face images are lost when nearest neighbor matrix is constructed, because face matrix is pulled into avector to calculate the weight matrix. Compared with Locality preserving projection (LPP),the weight matrix of neighborhood preserving embedding is obtained by calculated ratherthan prior get which more able to represent the relation among data. In addition, if the faceimages are pulled into one-dimensional vectors, dimension will be very high and calculationis more difficult, therefore, in this paper,2D neighbor preserving embedding algorithm ispromoted to improve old method. the local information of the face is taken into account incalculating the neighborhood matrix, our direct use a two-dimensional matrix to represent theface image rather than a one-dimensional vectors, therefore not only the local feature of theface are maintain, but also the amount of computation is greatly reduced. In order to verifythe proposed method, we have done a lot of experiments on three most commonly used facedatabase Yale, ORL and AR, the experimental results show that, comparing withneighborhood preserving embedding method, the improved two-dimensional neighborhoodpreserving embedding method is not only have been greatly improved in the recognition rate,but also the running speed is significantly improved.