Research On Face Recognition Based On Local Graph Structure And Weber Local Descriptor

The technology of face recognition has made great development in recent years,some commercial face recognition systems are gradually entering the market.However,its application condition is very strict.At the same time,under the non-restraint conditions,it is easily to be affected by the illumination,posture variations and partial occlusion,which leads to the decline of the recognition performance.Aiming at this problem,based on the analysis of the existing algorithms and feature extraction,this paper makes full use of the advantages of Weber local descriptor(WLD)and local graph structure(LGS),and focuses on how to improve it to extract more robust face features under the illumination,posture and partial occlusion conditions,then we propose a novel face feature extraction method Weber local graph structure(WLGS).Finally we verify its performance by experiment.The main work and innovation of this paper are as follows:(1)Focusing on the problem that the isotropic WLD operator cannot totally distinct local textures in the cases that directional differences exist,an anisotropic WLD operator is proposed by introducing the angle parameter and scale parameter in order to solve the problem that the gray scale change information has not fully reflected in the local window,which isotropic WLD operator exists.(2)Focusing on the problem that the LGS operator is asymmetrical in its structure and contains redundant information among adjacent pixels,by improving the number of unequal pixels in the left and right neighborhood of the LGS operator,this paper presents a symmetric LGS operator and proposes a more balanced approach to extract the texture information between pixels.(3)The WLD operator consists of two parts: differential excitation and gradient direction.On its basis,anisotropic WLD differential excitation and symmetric LGS operator are applied so that the differential excitation and gradient direction in the original WLD operator can be replaced,this paper proposes a new operator based on the Weber local graph structure(WLGS),which is an improvement of WLD operator,can extract more texture details and multiple direction gradient information.In this paper,the experiments are carried out on CMUPIE,YALE,FERET and LFW face databases,and compared with some classical feature extraction methods.It is obvious that the method achieve the better performance under the condition of the changes illumination,posture and partial occlusion.