Research Of Face Recognition Algorithm Based On Linear Subspace And Gabor Wavelet

Face recognition is an important part and a hotspot of pattern recognition and imageprocessing, and it is also an active subject in the area of biometrical recognition technology.Having the advantages of real-time, accuracy and non-contact, the technology of facerecognition is easier for user to accept. So it has a wide application in access-control andchecking-in system, etc.The topic of this thesis is the algorithm of face recognition, Feature extraction andrelated issues are discussed in depth. The main points are as follows:(1) Feature extraction is one of the most significant parts of face recognition systemwhich is summarized firstly. Principal Component Analysis(PCA) and Fisher LinearDiscriminant(FLD) are studied intensively so far which are wide applied and the main waysamong the subspace. However, these two methods can not represent the local feature of faceimage effectively, so Modular PCA, Modified Modular PCA and Modular FLD are proposedto improve them and enhance the recognition rate.(2) Owing to the advantage of extracting textural feature from image, Gabor wavelettransform is introduced to dispose face image. Because of too high dimension and a mass ofredundant information, Gabor textural feature is fused by using the average fusion method tolower the dimension of it preliminarily and get complete margin-message of image.(3) The fusion algorithms of Modular PCA, Modified Modular PCA and Modular FLDwith Gabor wavelet transform are proposed respectively. It can make full use of thepredominance of Modular PCA, Modified Modular PCA and Modular FLD to lower thedimension of Gabor feature. Lastly, Nearest Neighbor Classifier is adopted to identify andrecognize the face feature. Have an experiment and analyse the result which applies theproposed algorithms into ORL and YALE face database. The result indicates that the proposedalgorithms have a good performance in face recognition.