Font Size: a A A

Research On Multiple Pose Face Recognition Based On Improved LBP And Face Energe Image

Posted on:2016-03-17Degree:MasterType:Thesis
Country:ChinaCandidate:J Y HuFull Text:PDF
GTID:2348330542974001Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
In recent years,face recognition technology has been the research focus in the field of the pattern recognition and artificial intelligence technology,as a kind of very important biological characteristics,face has the advantages of non-contact,quickness,convenience and uniqueness,but face images have some change in a different environment(such as posture,light,shade,etc),the changing pose is one of the main factors hindering the development of the face recognition,when the face posture change is bigger,it will lead to recognition rate decline and can't even recognize.This paper centers on the key technology of the multi-pose face recognition,mainly to do the following work and research:(1)General narrates the background and significance in the research of face recognition,this paper expounds the present situation and the main technical method of the study on face detection and face recognition,illustrates the technical difficulties.(2)Introduce the establishment process of the face recognition system,expound the unique advantages of faces compared with other biometric methods,To the problem of the face images collected are often lower quality and affect the identification effect,Proposed three methods of image preprocessing in this paper,such as size normalized?image gray scale and histogram equalization.(3)Introduce the principle of Adaboost algorithm in detail,contain the concepts of rectangle features and integral figure,and elaborate the establishment method of the weak classifier,strong classifier and the cascade classifier respectively.The method based on the Haar-like characteristics of Adaboost algorithm for frontal face detection effect is better,but it is not robust for the multi-pose images.And haar feature has higher dimension,detection speed is slower,this pqper adopts the method based on the LBP characteristics of Adaboost algorithm,and train multi-pose face classifier.The detection speed is faster of the method of LBP characteristics,and the recognition effect is better for multi-pose face images.(4)Introduce some feature extraction algorithm,such as multi-scale local binary pattern,center-Symmetric local binary pattern,multi-scale block center-symmetric local binary pattern,because of the improved local binary pattern has a unique advantage,a large number of experimental results show that the texture characterization method can be used to research mutli-pose face recognition in this paper,and it has a certain robustness to multi-pose face image.(5)The multi-pose face recognition methods based on improved LBP features and faceenergy images are proposed,interducing the kinds of the face emerge image,the definition and buliding method of it is given.To extract the LBP feature of the face emerge and waiting recognition face,then determine according to the Hamming distance,and obtain the good recognition result.Finally,basing on the research of this paper,a multi-pose face recognition system is developed.
Keywords/Search Tags:multi-pose face recognition, local binary pattern, face mean energy image, face variance energy image, feature extraction
PDF Full Text Request
Related items