Research On Image Texture Classification Based On Improved LBP Operator

Texture is an important visual underlying feature and directly reflects the inherent properties of the object surface,it plays an important role in the field of image analysis and understanding.As a typical local texture descriptor,the local binary pattern is widely applied to machine vision and pattern recognition in various fields.However,the existing LBP algorithms are still suffering from the sensitivity of illumination change and texture rotation variation,as well as to the lack of ability to describe texture features of noisy image.Therefore,the LBP algorithms of improving the anti-noise ability and enhancing the robustness to image illumination and rotation variation are deeply studied in this paper.The main contents are as follows:?1?As one of LBP variants,Pairwise Rotation Invariant Co-occurrence Local Binary Pattern algorithm has poor robustness to image illumination and rotation variation,and has a high computing feature dimensionality.To deal with this problem,an efficient texture feature named enhanced pairwise rotation invariant co-occurrence extended local binary pattern which could fuse a variety of local texture structure information is proposed in this paper.First of all,the binary coding sequence are obtained by performing binary quantization on the neighborhood pixel gray value of each pixel point,the LBP value corresponding to the different neighborhood points of each pixel are used as the initial point of coding obtained by continuously rotating the binary coding sequence;And then,two co-occurrence directional vectors with rotation invariance are determined by using the central pixel point and the neighborhood initial points of coding corresponding to the maximum and minimum LBP values of each pixel respectively,and two spatial context co-occurrence pixel points at different scales are chosen along the two directional vectors on two different scale gray images;Next,the correlation information between the central pixel gray level feature,the neighborhood pixel gray level feature and the radial gray level difference feature of spatial context co-occurrence points are extracted by using the rotation invariant uniform descriptor of extended local binary pattern?ELBP?algorithm,and the texture structure of complex image is descripted by cascading the ELBP feature of each spatial context co-occurrence point;At last,Chi-square kernel support vector machine which trained by texture feature histograms of spatial context co-occurrence pixel pairs is used to complete the detection of image texture categories.Under the same experimental conditions,compared with the original PRICoLBP algorithm,the classification recognition rate of the proposed method was improved by 0.32%,0.57%,5.62%,3.34%,2.1%and 4.75%on the Brodatz?Outex?TC10?TC12??Outex?TC14??CUReT?KTH-TIPS and UIUC texture database respectively,The experimental results show that the improved algorithm is more robust to texture rotation variation and illumination change than a number of recent state-of-the-art LBP variant algorithms under the same conditions.?2?In order to solve the problem that the local binary pattern algorithm is sensitive to noise,a multi-scale co-occurrence median noise resistant completed local binary pattern?MSCoMNRCLBP?algorithm is proposed.First,according to the number of“1”in the binary sequence,the feature patterns whose bitwises 0/1 change are 4,6,or 8 times in the sequence are reclassified to form an extended rotation invariant uniform LBP(LBP^eriu2).Then,the gray value of a single pixel was replaced by the median value of the gray in the rectangular neighborhood of the pixel,and the CLBP operators is used to extract gray-scale difference sign,magnitude and center pixel gray-scale information between adjacent pixels at different scales.Meanwhile,the multi-scale co-occurrence strategy is used to extract the correlation information between the gray-scale difference sign,magnitude feature at different scales and the gray-scale feature of the center pixel respectively,and the co-occurrence features at different scales are concatenated to describe the image texture structure.Finally,the principal component analysis?PCA?was used to reduce the dimension of the above features,and the feature vectors of dimensionality reduction are sent to support vector machine to complete the texture classification.The proposed algorithm had higher recognition accuracy than other existing LBP variants through the noise free classification experiments on each standard texture library.Extensive experimental results show that the improved MSCoMNRCLBP algorithm not only has stronger noise resistance than other existing LBP variants,but also has stronger robustness to image illumination,rotation and scale variation.