基于特征融合与低秩分解的织物疵点检测  

Fabric Defect Detection Based on Feature Fusion and Low-rank Decomposition

作　　者：刘洲峰[1] 闫磊[1] 李春雷[1] 董燕[1] 王宝瑞[1] 

机构地区：[1]中原工学院,河南郑州451191

出　　处：《棉纺织技术》2017年第10期1-4,共4页Cotton Textile Technology

基　　金：国家自然科学基金资助项目(61379113);郑州市科技领军人才项目(131PLJRC643);河南省高校科技创新人才项目(17HASTIT019)

摘　　要：研究基于特征融合与低秩分解的织物疵点检测算法。采用超像素分割方法将待测图像分割为超像素块;分别提取各块灰度和HOG特征,构造融合特征矩阵;利用低秩分解方法将融合特征分解为低秩背景和显著疵点,依据显著度大小得到疵点显著图;最后选择最大熵阈值分割方法对显著图分割,得到检测结果。选取TILDA标准织物图像库验证算法有效性。结果表明:提出的算法能有效检测出织物疵点所在位置和形状。认为:本文提出的算法自适应能力较强,适用较多疵点类型,具有较高疵点检出率。Fabric defect detection algorithm based on feature fusion and low-rank decomposition was re- searched. The original image was segmented to superpixel block by superpixel segmentation method. Gray feature and HOG feature of each pixel block was extracted respectively for building the fusion feature matrix. Fusion fea- ture was segmented to low-rank background and salient defect by low-rank decomposition method. The defect sa- liency image was got based on the size of salience degree. Finally maximum entropy segmentation method was chosen to segment the saliency image and the detection result was got. TILDA standard fabric image database was chosen to test the validity of the algorithm. The results show that the suggested algorithm can detect the position and shape of fabric defect effectively. It is considered that the suggested algorithm has better self-adaptive ability. It is suitable for more defect types and has higher defect detection rate.

关 键 词：织物疵点 特征融合 低秩分解 疵点检测 显著图分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TS101.97[自动化与计算机技术—计算机科学与技术]