Research On Texture Classification Algorithm Based On Improved Local Binary Pattern

As one of the key features of images,texture features have been widely concerned in the field of image processing.Texture feature extraction technology is widely used in texture classification,face recognition,image retrieval and other fields.Local Binary Pattern（LBP）,as a kind of image feature extraction algorithm,has gradually become one of the main methods of texture feature extraction due to its extremely low computational complexity and strong texture discrimination ability.Since the LBP operator was proposed,various LBP improved operators have been proposed successively,which solve some of the shortcomings of LBP from different perspectives,thereby improving the accuracy of texture classification.But these improved operators greatly increase the computational complexity,and low computational complexity is one of the main reasons why LBP operators are widely used.For this problem,this paper studies LBP and its improved operator.The main research contents and innovations are as follows:（1）Three problems of the LBP operator are analyzed in detail:1)Discarding the magnitude information,2)Adopting fixed weights,3)Discarding the absolute gray level information of pixels,and the second problem is proposed for the first time in this paper.For the problem of discarding magnitude information,this paper innovatively proposes a method based on information entropy to quantitatively compare the proportional relationship between symbols and magnitude information in local pixels.For the last two questions,this paper presents examples to demonstrate.（2）For the first two problems,this paper combines the magnitude information with the weight,and proposes a Local Binary Pattern based on Magnitude Ranking(LBP^mr).For the third problem,this paper proposes the Global N-nary Pattern（GNP）operator by extending the CLBP＿C operator.The experimental results show that the texture classification accuracy of LBP^mr/GNP operator is only about 1.5%lower than that of the best LBP improved operator in recent years,but its computational complexity is lower than that of the best LBP.The improved operator is dozens of times lower than that of the the best LBP improved operator.（3）In view of the problem that the feature dimension of the LBP^mr operator is too large and the effect of the dimension reduction algorithm is not good,this paper improves the traditional feature dimension reduction method based on the dominant pattern,and applies the improved algorithm to the LBP^mr operator,which can effectively improve the density of texture information in LBP^mr feature images without adding additional computational complexity.Experiments show that the LBP^mroperator is better than the original feature dimension reduction algorithm based on the dominant pattern after applying the feature dimension reduction algorithm,and its texture classification accuracy is further improved.