基于多特征融合的织物瑕疵检测研究

Fabric defect detection based on multi-feature fusion

织物瑕疵纹理特征复杂,单一特征不能很好地反映纹理信息。为此,本文提出一种基于局部二进制模式(Local Binary Pattern,LBP)算子和灰度共生矩阵(Gray Level Co-occurrence Matrix,GLCM)的多特征融合算法。首先,对LBP算子进行了改进,提出一种基于邻域像素中值的中心对称LBP算子;然后,将其提取出的纹理特征和灰度共生矩阵提取的纹理特征进行融合;最后,通过极速学习机和支持向量机做分类实验,验证融合特征描述织物瑕疵纹理特征的能力。实验表明,本文方法提高了织物物疵点检测率,并且具有很好的抗干扰能力。

In order to solve the problem that single feature dose not well reflect fabric defects texture which more complexity than general, this paper presents a feature fusion algorithm which fuses Local Binary Pattern feature （LBP）and Gray level cooccurrence feature matrix （GLCM）. Frisfly, a new algorithm named MCS_LBP will be put forward based on LBP. Secondly, the texture features extracted by MCS LBP will mix features which are extracted by GLCM. Finally, the Extreme Learning Machine （ELM）and Support Vector Machine （SVM） are applied to do classification experiments so that we can test the ability of fusion characterization texture features. The results show that the tactics in this paper can improve methods herein woven material defect detection rate, and they have good robustness.