Fabric Defect Detection Method Based On Statistical Characteristic

In the textile automatic production, defect detection is a key mean to ensure the quality of products. Compared with the traditional manual detection, computer image processing technology has the advantage of high precision and high efficiency in fabric detection. In the process of testing, the accuracy of the texture feature expression has important effect on the performance of the fabric defect detection algorithm. In this research, we studied the application of fabric texture statistical characteristic in automatic defect detection.The main contents are as follows:(1) According to RAMF and GGCM, we proposed a new fabric defect detection method. The GGCM matrix builds a second-order statistical feature with gray level and gradient information. Combining the RAMF, this algorithm can improve the separation of defect part and background part and the defect detection rate. The experimental results verified the effectiveness of this method.(2) A new defects detection method with the nature of anti-noise and multi-scale was proposed. Because the MB_LBP operator has the limitation on scale expansion of the square neighborhood, we use the BRINT_LBP operator to extract the texture characteristics of fabric. BRINT_LBP operator can extract image texture feature on the circular neighborhood under different scales, and has robustness to the noise. We construct a multi-scale BRINT_LBP feature by cascade the single scale BRINT_LBP feature and accomplish the testing combing the SVM classifier. A large number of simulation experimental results show that this method improves the textile defect detection accuracy, and can resist noise to a certain level.This research has positive effects on promoting the development of fabric defect automatic detecting algorithm.