Fabric Defect Detection Using Modified Local Neighborhood Analysis

Abstract

AbstractFabric defect detection plays a crucial role in the textile industry to improve the quality of service of the fabric texture. Automatic fault detection in fabric is challenging because of the variety of texture patterns, manufacturing defects, defects due to dyeing, and defects due to external environmental conditions. Existing local neighborhood analysis (LNA) for defect detection has given poor performance for smaller and light color variation defects. To deal with such conditions, this paper presents the unsupervised modified local neighborhood analysis (MLNA) for finding defect in non-patterned fabric. The threshold value used for detection of defect depends upon mean, standard deviation, and entropy of local homogeneity measure. The performance of the system is evaluated on the in-house database based on the percentage defect detection rate. The results of the proposed method are compared with previous methods such as wavelet transform and Gabor transform, and it is observed that the proposed method detects 97.33% of defects and this is much better than the detection rates of LNAs and other existing methods.KeywordsModified local neighborhood analysisLocal homogeneity measureFabric defect detectionNon-patterned fabric