Research On Face Recognition Based On Improved LBP Features

Face recognition is typical problems for image pattern analysis,understanding and classification compute, which has important theoretical research value and broad application prospects. In recent years,based on local patterns and texture features,face recognition has achieved remarkable success,the local modeling method which against the light,post change and local deformation has become one of the mainstream ways.The method of face detection based on local binary pattern has achieved excellence effectiveness,but the discrimination is decline prominence in the condition of larger illumination or complex image-forming condition. In this thesis,the LBP features space is studied and analyzed from the texture features of images,and directed towards the shortcoming of LBP features in the process of face recognition, a series of improved resolutions are presented,the main works as follows:1,In connection with the problems of large illumination,the gamma correction difference of gaussian (DoG) Filtering and contrast equalization are utilized to reduce the influence result from the extreme image-forming conditions.2,By means of the definition of LBP,a improved local ternary pattern for extraction of textures features is presented. LTP has added -1 valued quantization threshold coding which becomes more robust and resistance to illumination and noise,also inherits most of the other advantages of LBP,to a great degree, enhanced the discriminate of local texture features.3,Against the complex dimensions,and the vectors of optimal classification are presented,in this paper,the Fisherface algorithm which combined PCA with Fisher Linear Discriminant are used to reduce the dimensionality and optimize discriminative classification. 4,Illumination-preprocess,LBP/LTP features and Fisherface algorithm which are combined to test and anlyze on the standard face databases so as to prove the progress and strengthen of improved LBP features,promising results of experiment are indicated that these methods are valid,accuracy and feasible.