Research And Its Application Of Image Texture Classification Based On Contourlet Transform And Local Binary Pattern

With the spread of computer technology and artificial intelligence,there has a very wide range of applications in the image,such as image feature extraction which dominates the important role and is also one of the most hot research topic in the field of computer vision currently.In the past,the deeply study of local characteristics promoted the rapid development in the all field of computer vision,one of the most famous local binary pattern(LBP)is a very simple and efficient local feature descriptor.Since LBP algorithm was proposed,it is attracted by many researchers.The improved algorithm was applied to many fields of computer vision and pattern recognition,including texture classification,face recognition,target detection,etc.But all of LBP algorithms still exist some shortcomings and bottlenecks,such as too long in the length of histogram vector,lack of outstanding ability in LBP rotation invariant,not strong enough in noise robustness.In order to enhance identification ability of the LBP and improve robustness against noise,after the key technologies in texture classification which is used in this paper are studied,we propose some new algorithms and applications.In addition,the Contourlet transform,which can extract the image intrinsic geometric features,inherits the anisotropic and multi-scale relations of Curvlet transform,therefore the paper also proposes the method of texture classification based on Contourlet transform,and applies to the forgery note recognition.In this paper,the main work and contributions are as follows:1.Many famous texture feature extraction algorithm is studied in texture classification,such as LBP,CLBP,DLBP,SCLBP,etc.And the detailed working principles are introduced,it analyzes all of their advantages and the aspect of improvement,introduces their application in the typical areas.2.The improved algorithm based on local binary pattern is studied in this paper.Due to the weakness of identification capability and antinoise ability in LBP,on the basis of previous studies,this paper is to improve the BRINT recognition algorithm,and propose a new scale invariant feature extraction method based on BRINT.Through the analysis of the scale invariant space in BRINT,we can obtain the optimal scale invariant features,and getbetter result applying it into different texture database.First,LBP descriptor solves the problem of rotation invariant and gray invariant,but do not perform very well in scale-invariant features.We put forward the method to solve the bottleneck.Second,it is different from other traditional local scale-invariant features,we do not need to estimate the local scale,only use the global descriptor to achieve scale invariant features.Finally,we utilize histogram across different scale space,to maximize each component finally,and integrated into an optimal histogram,realize scale invariant features.The new scheme can get certain promotion in the texture database.Forgery note texture classification based on contourlet transform and statistical feature are studied.First,introduce the basic situation of notes in identification,in detail introduce the operating principle of contourlet transform as well as relevant variant.Then use frequency subband decomposition from contourlet transform into statistical feature extraction,Combining the method of gray level co-occurrence matrix,realize the note texture classification based on support vector machine(SVM).Finally compared with the related methods,we verify the feasibility of the algorithm.