基于窗阈值局部二值模式的织物疵点检测算法  

Fabric defect detection based on window threshold local binary patterns

作　　者：付蓉[1,2] 石美红[1] 徐步高[3] 

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048 [2]西安电子科技大学电子工程学院,陕西西安710075 [3]德克萨斯大学奥斯汀分校人类生态系,美国德克萨斯州78712

出　　处：《计算机集成制造系统》2010年第9期2009-2015,共7页Computer Integrated Manufacturing Systems

基　　金：陕西省科技厅13115科技创新工程资助项目(2008ZDKG-36);陕西省教育科技资助项目(05JC13);西安工程大学校管课题资助项目(2010XG15)~~

摘　　要：为准确描述不同织物的纹理结构,提出一种改进的局部二值模式,为不同纹理特征创建了相应的主要概率模式子集。在该特征提取算法的基础上设计了一种基于窗阈值的织物疵点检测算法,并对无图案和有图案织物分别设置了参数。该算法首先使用自适应局部二值模式获取无疵点织物图像特征并确定疵点判断阈值,然后将待检测织物图像分割为大小相同的检测窗,并提取同样特征与阈值进行比较,以判断该窗是否为疵点窗。对无图案和有图案织物的参数分别进行了讨论分析,以获得精确的分割结果。实验证明,所提出算法的疵点检测结果在视觉上更加细腻、误检率更低。To describe different texture structure accurately,an improved local binary patterns method was proposed. Corresponding main pattern sets for different texture structure were established by this method. Based on the proposed method,an effective threshold-based fabric defect detection algorithm was designed. Parameters were set up for patterned and unpatented fabric. Firstly the features of free defect image were extracted by Adaptive Local Binary Patterns（ALBP） and the defect judgement threshold was obtained. Then the image to be tested was divided into same size detection windows from which ALBP features were also extracted. The features were compared to the threshold to find the defective window. To obtain precise segmentation results,the parameters of patterned and unpatented fabric were discussed respectively.The experiments showed the detection effect of the proposed method was comparatively better than traditional local binary patterns from visual aspect and detection accuracy.

关 键 词：局部二值模式 图像分割 疵点检测 工业检测 织物 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]