Image Texture Feature Extraction Method And Application Based On LBP

Image classification is an important research direction in the field of computer vision,mainly achieved through distinguishing semantic information in images to recognize different categories of images.In the process of image classification,image feature extraction is a crucial step,and texture feature is an important type of image feature.Local Binary Pattern(LBP)is a texture feature extraction algorithm with significant discriminability and robustness.However,the multi-scale LBP algorithm has the problem of missing neighborhood point information,and the fixedradius LBP algorithm only utilizes the relationship of single pixel points,making it difficult to describe the central point gradient information.Therefore,this research aims to enhance the feature extraction ability of LBP for images,and the main research contents are as follows:(1)Based on neighborhood gradient multi-scale LBP feature extraction method.Aiming at the problem that multi-scale LBP cannot fully utilize neighborhood point information,leading to the loss of some critical texture features,a new algorithm based on neighborhood gradient multi-scale LBP feature extraction is proposed.This algorithm characterizes the gradient changes between multi-scale neighborhood point information by defining the neighborhood gradient descriptor and mean gradient descriptor and constructs a new threshold to obtain a more accurate data representation.By adding the initial uniform local binary pattern gradient descriptor,the features of the four descriptors are linearly connected to fully utilize multi-scale neighborhood information and extract texture features on a larger scale,improving the feature extraction ability of the algorithm.The classification experiment on UIUC,Kylberg,KTHTIPS2 b,and Outex＿TC＿00012 texture datasets shows that the algorithm improves the feature extraction ability with relatively less runtime consumption,achieving higher classification accuracy.(2)Research on feature extraction method based on multi-pixel LBP.Aiming at the problem that LBP only considers the difference relationship between individual pixels,making it difficult to describe the central point gradient changes and fully extract image texture features,a feature extraction algorithm based on multi-pixel LBP is proposed.This algorithm uses three multi-pixel descriptors(central point descriptor,neighborhood point descriptor,central point and neighborhood point descriptor)to generate three feature maps,reflecting gradient information in different directions and distances and extracting texture features.After the statistical feature vector,the features of the three descriptors are linearly combined into a complete texture feature vector.Experiments on the texture data of trees and stones such as UIUC,Kylberg and KTH-TIPS2b show that,on the basis of expanding the extraction range of texture features,this algorithm can better extract important texture features by describing the gradient changes of center points in more detail,and further improve the accuracy of image classification.(3)A texture classification prototype system based on neighborhood gradient multi-scale LBP is designed and implemented.Firstly,the functional modules and processes of the system are introduced in detail,and then the key technologies of the system are introduced.Finally,the results of experimental cases show that the system is highly feasible.