Research On Image Feature Description Method Of Illumination Invariance

Image feature extraction technology is one of the main research topics in computer vision.It has been widely applied to bio-recognition,image retrieval,target detection and more fields.Variations of illumination present a great challenge to image recognition in the real application,including insufficient illumination,overexposure and shadow,which make it difficult to obtain image features from images with great discrimination and strong robustness.Aiming at the drawbacks that traditional feature extraction methods have poor discriminant ability and inadequate texture description under complex illumination conditions,this thesis proposes several improved methods.Extensive experiments are conducted on some public databases to verify the effectiveness of the proposed algorithms.The main work and contributions are summarized as follows:(1)In order to solve the problems of LGP operator with single-layer model,Weighted Local Synergistic Gradient Pattern(WLSGP)is proposed.WLSGP operator presents a twolayer structure model and a weight coefficient allocation model,which can extract more detailed texture information,to acquire better robustness feature under complex illumination conditions.(2)In order to solve the problem that traditional LBP operator lose the co-occurrence of adjacent local pattern,Local Binary Co-occurrence Gradient Pattern(LBCGP)is proposed.LBCGP operator expresses the co-occurrence between adjacent pixels when the code is 0 or 1 at the same time,and describes more texture structure information with stronger discriminability.(3)In order to solve the problem that the differential excitation component of WLD operator fails to make full use of the local texture information in detail,Differential Synergistic Excitation Pattern(DSEP)is introduced.Focusing on the drawback that WLD operator cannot extract gray information adequately using isotropic LOG operator,the LOG operator with variable-scale and variable-angle is introduced into differential cooperative excitation,and Anisotropic Differential Synergistic Excitation Pattern(ADSEP)is proposed.(4)In order to solve the problem that WLD operator has limitations in feature description,this paper combines the advantages of ADSEP operator and WLSGP operator,Anisotropic Weber Synergistic Gradient Descriptor(AWSGD)is proposed.The AWSDG operator replaces the differential excitation component and directional component in the original WLD with ADSEP operator and WLSGP operator respectively,the algorithm can extract richer texture details and direction gradient information.Finally,we use the XGBoost classifier to conduct related experiments on face databases CUMPIE,Yale B and texture databases PhoTex,RawFooT.The proposed algorithms are compared with the state-of-art feature description algorithms.The experimental results show that the proposed operators are robust to illumination changes and have a high recognition rate.In a word,the algorithms proposed in this paper have certain theoretical significance and practical value in the field of image recognition with varying illumination.