Neighborhood Difference Based Texture Image Classification Methods

Texture classification is an important research topic in the field of computer vision and pattern recognition.Texture classification methods have been successfully applied to many fields.In recent decades,various feature extraction methods have been proposed for texture classification.In general,there are two broad categories,the spatial domain methods and the spectrum domain methods.The spatial domain methods are used to deal with the relationship between image pixels and their neighborhoods.The spectrum domain methods utilize transformation coefficients.Image classifications still have the following problems.The texture features are high in dimensions,and rotation,illumination,scale changes,viewing angle changes,noise,etc.are not robust.Aiming at the above problems,the paper proposes two texture classification methods.Jumping and refined local pattern for texture classification includes jumping local difference pattern and a refined complete local binary pattern.Jumping local difference pattern includes the jumping information in the local region,extracts the second-order difference count feature between the neighboring pixels in the local region,and the diagonal directional feature based on the specified distance,and is used to suppress the influence of noise and rotation on the image classification.In order to capture the detailed information left by jumping local difference pattern,a refined complete local binary pattern is proposed,which is to re-divide the non-uniform mode difference features between the central pixel and its neighboring pixels,and extract the microstructure information and macro structure information from image.Moreover,the feature dimension of the method is low and the calculation amount is small.Locally directional and extremal pattern for texture classification includes direction information and pixel intensity information of the image texture.The method consists of local direction difference count pattern and neighborhood extreme local pattern.The former extracts difference information in different directions from the odd neighborhood and the even neighborhood of the central pixel.The latter extracts the position information?maximum value,minimum value?of the extremum in the local region,the rotation invariant unified pattern information between the extremum and the neighboring pixel points in the local region,and introduces the residual model to obtain the center pixel compression information based on extremum.Experiments on the internationally recognized standard texture databases?Brodatz,CUReT,UIUC,VisTex,Kylberg,Kth-tips2-a,Outex_TC₀0010,DTD,Prague,Stex,and Kth-tips2-a?demonstrate that the two proposed methods in this paper compared with the other six representative texture classification methods are robust to different imaging conditions,such as rotation,illumination,scale change,viewing angle change,noise,etc.,and can obtain more satisfactory classification.performance.