尺度选择完备局部导数模式及其在热轧带钢图像分类中的应用研究  

Scale selective completed local derivative pattern and its application in hot rolled strip image classification

作　　者：梁纬 逯洋[1] 王淳 张桂杰[1] Liang Wei;Lu Yang;Wang Chun;Zhang Guijie(College of Mathematics and Computer,Jilin Normal University,Siping,136000,China)

机构地区：[1]吉林师范大学数学与计算机学院,四平136000

出　　处：《南京大学学报（自然科学版）》2022年第4期615-628,共14页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(61572239,61976106);吉林省发展和改革委员会创新项目(2021C038-7)。

摘　　要：纹理特征提取是纹理分类中最关键的一步,由于成像条件的不可预知,纹理图像中存在旋转、尺度、噪声等各种因素的变化,给纹理分类的研究工作带来了挑战.为了增强纹理特征提取算法对旋转、尺度和噪声变化的鲁棒性,提出尺度选择完备局部导数模式(Scale Selective Completed Local Derivative Pattern,SSCLDP).首先,采用自适应中值滤波器对图像进行降噪处理.其次,采用二维高斯滤波器生成该图像的尺度空间.在每个尺度下使用完备局部导数模式(Completed Local Derivative Pattern,CLDP)提取该图像的旋转不变特征,跨尺度取最大值作为该图像的尺度不变特征.将SSCLDP与同类算法在七个公共纹理数据集和热轧带钢图像数据集上进行了实验,实验结果表明,SSCLDP在纹理图像分类和热轧带钢图像分类上有较好的工程应用价值.Texture feature extraction is the most important step in texture classification. Due to the unpredictable imaging conditions,there are many changes in texture images such as rotation,scale and noise,which bring challenges to the research work of texture classification. The SSCLDP(Scale Selective Completed Local Derivative Pattern) is proposed to enhance the robustness of texture feature extraction algorithm against rotation,scale and noise changes. Firstly,an adaptive median filter is used to suppress noise in texture images. Secondly,a two-dimensional Gaussian filter is used to generate the scale space of the image. At each scale,the CLDP(Completed Local Derivative Pattern) is used to extract the rotation-invariant feature of the image,and the maximum value across scales is chosen as the scale invariant feature of the image. Experiments are carried out on seven public texture datasets and the hot rolled strip image dataset with SSCLDP and other state-of-the-art algorithms.Experimental results show that SSCLDP has good engineering application value in texture image classification and hot rolled strip image classification.

关 键 词：局部二值模式 旋转不变 尺度不变 对噪声鲁棒 纹理分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]