The Research Of Face Recognition Algorithm On2DPCA With Supervised LPP

Because of biological technology the Face Recognition technology had developedvery fast in recent years. Face recognition has been widely used in various fields, suchas entrance guard system, security systems, Crime screening, the financial sector. Facerecognition is known as a branch of Applied Science of pattern recognition. It has aconsiderable advantage Compared with other identification system such as fingerprintrecognition, iris recognition system, gene identification system, retina identificationsystem. The advantage are easy to use, is not easy to be lost, not aggressive and the useof the image characteristics protect the privacy. When these advantages combined withartificial intelligence they will explosive very strong energy. Many people pay theirpassion on the research and development, its makes them developed rapidly. After thestudy of traditional literature and methods, this paper presents a new method to improvethe face recognition rate, so it can also have good effect in some more complex case.In this paper, we combined the traditional2DPCA and LPP algorithm. Since LPP isan unsupervised learning method, it ignored some kind of internal information. We use asupervised learning approach to complete the learning process that can provide us moreuseful information for recognition. First of all, we reduce the dimension of face imagesin ORL face database, there are many methods of dimensionality reduction, such as thetraditional2DPCA. But this method will need to make the face image one-dimensional,when the image is very large the amount of calculation is huge, in this paper we use2DPCA to reduce dimension, this can save a lot of computation time. Whendimensionality reduction we discard some eigenvectors whose Eigen values are zeroand remain other eigenvectors. Then we remain the most useful information we need.After the image dimensionality reduction, we use LPP to get the matrix of a lineartransformation of training images. Because the LPP method does not use categoryinformation in computing base vector, this paper fully considers the categoryinformation when calculating the basis vectors. We discarded these eigenvectors that notreaction difference of the people mainly retained the vectors which reaction difference of the people that are useful to the recognition rate. The test proves that this algorithmhave higher recognition rate compared with other algorithms such as:2DPCA, LPP,LDA. This method also provides a certain reference for the research on facerecognition.