Research On Texture Image Feature Extraction And Classification Of Based On Improved LBP

Texture is a basic feature of the surface appearance of almost all-natural objects and is ubiquitous in nature.Texture,as an important feature and visual cue of many types of images,plays a key role in various applications of computer vision and image analysis.In order to extract effective texture features,many texture descriptors are proposed.Among them,local binary pattern(LBP)and its variations are one of the most widely used texture descriptors.LBP descriptor is simple,easy to understand,and has the advantages of high efficiency of feature computing,strong feature discrimination,and low computational complexity,etc.,it has been widely used in the field of computer vision.However,traditional LBP only extracts texture information from the neighborhood of a single point pixel without considering the local region and global texture information,which leads to the unsatisfactory accuracy of texture classification.This paper focuses on the basic theory of LBP and local binary pattern variance(LBPV).In order to overcome the defects of traditional LBP and solve the fusion problem of local texture information and global texture information,this paper refers to the advantages of LBPV in considering both local texture information and global texture information,introduces the adaptive weight based on sampling radius,and proposes a new solution for more comprehensive description of texture information.Firstly,aiming at the problem that traditional LBP does not consider local region and global texture information,this paper focuses on the basic theory of LBPV.On this basis,in order to more accurately express local texture features,this paper uses the square of local variance as the histogram cumulative weight,and proposes a new local binary mode square of variance(LBPV~2)descriptors.Simulation results show that LBPV~2 can improve texture classification performance to some extent.Secondly,since the traditional LBP only encodes the neighborhood in a single scale,and its feature expression ability is too simple,this paper focuses on the study of the joint multi-scale local binary pattern(JLBP).Although this method can fuse multi-scale texture information with the same feature dimension,its fusion scheme does not take into account the amount of local region information.In order to solve this problem,this paper proposes an adaptive weight joint multi-scale local binary pattern(AWJLBP).Experiments show that AWJLBP improves texture classification performance significantly compared with traditional LBP and JLBP.Finally,in order to obtain a more comprehensive and reasonable texture descriptor,this paper proposes a joint scale local binary pattern variance(JLBPV)texture descriptor.In order to further improve the performance of texture classification,this paper introduces local variance squared and adaptive weight scheme,and proposes the adaptive weight joint scale LBPV~2 descriptor(AWJLBPV~2).The comparison experiment shows that AWJLBPV~2 has obvious performance improvement compared with the traditional methods.