Font Size: a A A

Face Recognition Based On Locality Preserving Projections

Posted on:2017-02-27Degree:MasterType:Thesis
Country:ChinaCandidate:L L MeiFull Text:PDF
GTID:2348330503965768Subject:Applied Mathematics
Abstract/Summary:PDF Full Text Request
With the development of computer and science technology, a more convenient and efficient identification technology face recognition technology was proposed in such situation, and has become the most popular research topic in the field of pattern recognition and machine learning in recent years,and has been widely applied in the field of intelligent monitoring,public safety management and digital authentication.Generally, a face recognition system consist of four process: gathering face image, image preprocessing, feature extraction and face recognition.Among,feature extraction and matching is the key. This paper in view of the feature extraction and face recognition these two aspects are studied.1?Research the principal component analysis(PCA),linear discriminant analysis(LDA),local preserving projection(LPP) and other typical face feature extraction methods which include the one-dimensional feature extraction and the two-dimensional feature extraction two cases.For several kinds of feature extraction methods in each case, we analysis their differences and relations in order to improve the algorithm can bu found below.2 ? An face recognition algorithm called improved adaptive locality preserving projection is researched.In order to avoid the effects of recognition rate caused by the selection of parameters, Firstly, a parameter-free graph construction strategy is designed,which can adaptively choice neighbors of each sample point and determine corresponding edge weights.Then, because of high dimensionality problem in the matrix calculation process, we use QR decomposition to reduction dimension.Finally,conjugate orthogonalization is used to to reduce the statistical correlation between feature vectors and improve the recognition rate by ensuring that the projection axis having statistically uncorrelated. In the experimental aspect,firstly,we carry out the selection experiment of scale factor ?.Then, Experiments on several databases(ORL ? YALE) show that the method is effective and stable and has a higher correct recognition rate compared with the LPP,DLPP and LMMC algorithm.Finally, we add two comparative experiments,one is the comparative experiment of the algorithm in this paper using different classifiers,the other one is the comparative experiment of the algorithm in this paper with the uncorrelated condition and without the uncorrelated condition.Experimental results show that the algorithm is effective.3?A bi-directional compression face recognition method which combine 2DLPP and 2DPCA is researched.First of all,to overcome the shortcomings of one-dimensional feature extraction,we upgrade the one-dimensional LPP algorithm, and switch to2 DLPP algorithm. Second, various two-dimensional algorithms can only reduce the dimensions from one direction of target data for feature extraction and the choice of dropping the dimensions is restricted, so we adopt the method that reducing the dimensions from the horizontal and vertical directions of target data for feature extraction in order to improve the recognition rate.In the experimental aspect,firstly,in order to verify the advantage of the algorithm(2DLPP+2DPCA) in this paper in the aspect of recognition rate,we carry out the comparative experiment with other algorithms on ORL face database.Then, in order to verify the effectiveness of the algorithm in this paper(2DLPP+2DPCA),we carry out the comparative experiment on other face database(YALE) which is different with the ORL face database.Finally, in order to verify the correctness of the mentioned algorithm theory that we can improve the recognition rate when reducing the dimensions both in column direction and line direction in this paper,we carry out the comparative experiment between the2DLPP+2DPCA algorithm and the 2DPCA+2DLPP algorithm.
Keywords/Search Tags:Face recognition, Feature extraction, Locality preserving projections, Parameter-less, Reduction dimension
PDF Full Text Request
Related items