基于局部补数–导数模式的光照反转和旋转不变纹理表达  

Illumination-Inversion and Rotation Invariant Texture Representation Based on Local Complement and Derivative Pattern

作　　者：辛亮亮 宋铁成 张刚[1,2] 高陈强[1,2] 张天骐[1,2] XIN Liang-Liang;SONG Tie-Cheng;ZHANG Gang;GAO Chen-Qiang;ZHANG Tian-Qi(School of Communication and Information Engineer-ing,Chongqing University of Posts and Telecommunications,Chongqing 400065;Chongqing Key Laboratory of Signal and Information Processing,Chongqing University of Posts and Telecommunications,Chongqing 400065)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]重庆邮电大学信号与信息处理重庆市重点实验室,重庆400065

出　　处：《自动化学报》2021年第4期924-932,共9页Acta Automatica Sinica

基　　金：国家自然科学基金项目(61702065,61671095,61571071);信号与信息处理重庆市市级重点实验室建设项目(CSTC2009CA2003)资助。

摘　　要：针对现有局部二值模式(Local binary pattern,LBP)算法对光照反转变化敏感和特征描述力不足的问题,本文提出一种基于局部补数–导数模式(Local complement and derivative pattern,LCDP)的纹理表达方法.其中,局部补数模式(Local complement pattern,LCP)用于编码原始图像空间中的近邻差分符号信息,局部导数模式(Local derivative pattern,LDP)用于编码不同尺度下(一阶和二阶)高斯导数空间中的近邻差分幅值信息,二者对光照反转和图像旋转均具有鲁棒性.为实现对差分符号和差分幅值的联合统计,同时维持特征的紧致性,进一步提出基于均值采样的联合编码方案.最后,对联合编码的结果进行多尺度直方图特征表达.实验表明,该方法能够有效提高线性和非线性光照反转条件下纹理图像的分类精度.The existing local binary pattern(LBP)based algorithms are sensitive to inverse illumination changes and have limited ability for feature description.In view of this,a method for texture representation is proposed based on local complement and derivative pattern(LCDP).In LCDP,local complement pattern(LCP)encodes the signs of neighbor differences in original image space whereas local derivative pattern(LDP)encodes the magnitudes of neighbor differences in(the first and the second order)Gaussian derivative space at different scales.Both LCP and LDP are robust to inverse illuminations and image rotation.Furthermore,a joint encoding scheme based on mean sampling is proposed.This is used to establish the joint statistics of difference signs and difference magnitudes while remaining compact features.Finally,the texture descriptor is obtained by constructing multi-scale histograms of jointly encoded features.Experiments demonstrate that the proposed method can effectively improve the classification accuracy of texture images under both linear and nonlinear inverse illumination conditions.

关 键 词：纹理分类 特征提取 光照变化 局部二值模式 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]