王雪娇

[摘    要]针对纺织行业花式色纱纬编针织面料依据图案纹理自动分类问题，提出一种基于局部二值模式（Local binary pattern，LBP）和支持向量机（Support vector machine，SVM）的组合模型分类方法，对面料纹理进行特征提取和类别判定。对图片采用中值滤波去噪，通过优选采样模板半径和核函数，以LBP旋转不变模式提取面料纹理LBP特征直方图，并利用参数优化的SVM支持向量机的分类模型对面料进行分类。实验结果表明：采样模板半径为5像素点的LBP旋转不变式算法能较优地满足3类面料纹理的特征提取要求，经过参数优化的SVM分类模型，实现3类面料纹理分类的准确率达96.67%，分类效果较好。

[关键词]花式色纱面料纹理;图像处理;特征提取;自动分类

[中图分类号]TP391.4 [文献标志码]A [文章编号]2095–6487（2021）11–0–03

Automatic Extraction and Classification of Fabric Texture Based on LBP and SVM

Wang Xue-jiao

[Abstract]To solve the problem of automatic classification of fancy color yarn weft knitted fabric according to pattern and texture in textile industry， a combined model classification method based on Local binary pattern （LBP） and Support Vector Machine （SVM） was proposed. Feature extraction and classification determination of fabric texture were carried out. The image was denoised by median filtering， and the LBP feature histogram of fabric texture was extracted by LBP rotation invariant mode by optimizing sampling template radius and kernel function， and the three kinds of fabric were classified by SVM classification model with optimized parameters. The experimental results show that the LBP rotation invariant algorithm with a radius of 5 pixels of sampling template can better meet the feature extraction requirements of the three kinds of fabric texture. The classification accuracy of the three kinds of fabric texture is 96.67% after the SVM classification model with optimized parameters， and the classification effect is good.

[Keywords]fancy color yarn fabric texture; image processing; feature extraction; automatic classification

纺织面料具有明显的图案纹理属性且种类繁多，传统的面料图案纹理检索都需要通过人工在实物样品库中查找，不同图案纹理面料的分类检索为纺织企业增加了庞大的工作量。随着计算机技术的不断发展，图像处理技术为实现面料图案纹理的自动识别与分类检索提供了有效手段。面料图像纹理的分类检索已成为研究热点，高效地分类检索这些图像具有重要意義。

局部二值模式算法（Local Binary Patterns，简称LBP）是一种用来表达图像纹理特征的算法，该方法是1996年由芬兰奥卢大学的Timo Ojala等人。2002年Timo Ojala等人再次对LBP算法进行了改进，提出了具有旋转不变性的LBP改进算法，解决了图像旋转对LBP纹理特征的影响。LBP纹理特征提取算法广泛应用于图像处理和机器视觉等领域。……