A Hybrid Scheme for Online Detection and Classification of Textural Fabric Defects

Online automatic fabric defect detection and classification of the localized defect types are two vital stages in production line of textile manufactures. Here a hybrid approach is proposed for online detection of defects through serial fabric images and then classifying the localized defect types. First, defects are detected and localized by using a modified local binary pattern (LBP) operator and second, to characterize the defective regions, textons are utilized. Different classes of fabric defects locally cause different types of texture and therefore the classification of defects can be formulated as a texture classification problem. In the state-of-the-art texture analysis approaches a texture is characterized through textons describing local properties of textures. For the first time, in this paper the approach is used for classification of fabric defects. The employed dataset in this study is provided by fabric laboratory of University of Hong Kong. Images in the dot-patterned fabric database contain six types of well-known defects. Experimental results have yielded excellent results such that classification accuracy of detected defect types is determined 100%. The low computational complexity and high robustness of the proposed scheme confirm the usefulness of this approach for online fabric inspection.