Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (22): 251-255.DOI: 10.3778/j.issn.1002-8331.1908-0407

Previous Articles     Next Articles

Polarizer Visual Defect Detection and Classification Based on Improved LBP and SVM Algorithm

HUANG Guangjun, DENG Yuanlong   

  1. College of Mechatronics and Control Engineering, Shenzhen University, Shenzhen, Guangdong 518060, China
  • Online:2020-11-15 Published:2020-11-13



  1. 深圳大学 机电与控制工程学院,广东 深圳 518060


The visual defects of polarizer have a direct and crucial effect on the quality of TFT-LCD panels. In order to improve the recognition accuracy of polarizer defect image, the recognition method based on improved Local Binary Pattern(LBP) descriptor and Support Vector Machine(SVM) is proposed. Firstly, the defect image is obtained by using dark field imaging principle. For each defect image, it is divided into different regions, and the LBP features are extracted from each region to form high-dimensional composite features. The pixel mean features of different partitions are integrated with the LBP composite features. With the help of Principal Component Analysis(PCA), the correlation and noise between the features are eliminated. Then the Linear Discriminant Analysis(LDA) is used to further project and transform to low-dimensional features. Finally, the support vector machine is utilized to classify the above features. The advantages of improved LBP descriptor, PCA, LDA and SVM are combined, and simulation experiments are carried out in a total of 250 database. The results show that the classification accuracy by using the improved LBP is 99.2%, and the recognition time is 0.92 s, thus meeting the practical application requirement of industrial production line.

Key words: polarizer, defect detection, LBP descriptor, Support Vector Machine(SVM), Principal Component Analysis(PCA), Linear Discriminant Analysis(LDA)


偏光片的外观缺陷是影响TFT-LCD面板质量的重要因素之一。为提高偏光片外观缺陷图像识别的准确性,提出一种改进局部二值模式(Local Binary Pattern,LBP)描述符和支持向量机(Support Vector Machine,SVM)的识别方法。缺陷图像通过暗场成像原理获得,将缺陷图像划分为不同的区,对每一个区域提取LBP特征并组成高维复合特征。将不同分区的像素均值特征与LBP复合特征进行集成,利用主成分分析(Principal Component Analysis,PCA)消除特征间的相关性和噪声,使用线性判别分析(Linear Discriminant Analysis,LDA)进一步投影变换至低维特征,使用支持向量机对上述特征进行分类。结合改进LBP描述符、PCA、LDA和SVM四种算法的优点,在总数250的数据库中进行仿真实验,结果表明,该方法识别准确率达到99.2%,单张图像识别时间为0.92 s,完全满足工业生产线的实际应用要求。

关键词: 偏光片, 缺陷检测, LBP描述符, 支持向量机(SVM), 主成分分析(PCA), 线性判别分析(LDA)