Fabric defect detection based on completed local quartet patterns and majority decision algorithm

Abstract

Visual inspection is a main stage of quality control process in many applications. Surface defect inspection plays an important role in fabric textile quality control systems. Each type of abnormality, that disturb the repetitive texture of fabric is called defect. The main aim of this paper is to propose a rotation-invariant fabric defect detection approach with high detection rate in comparison with well-known methods. In this paper, a new version of local binary pattern, that is called completed local quartet patterns, is proposed to extract fabric image local texture features. In the train step, first, the size of fabric repetitive patterns is computed using autocorrelation function on a defect-less fabric image, the completed local quartet patterns operator is applied to whole of this image and to a set of non-overlapping windows. The maximum value of non-similarity amount between feature vector extracted from the whole image and each window is considered as the defect-less threshold. Considering the threshold obtained in the training the fabric defects are detected. In order to evaluate performance, a benchmark dataset of patterned fabric images is used, that includes three groups of fabric patterns and six defect types. Experimental results indicated that the proposed method can detect defects with 97.66 percent detection rate. Simple process, rotation invariant, gray-scale invariant and high detection rate are main advantages of the proposed approach. Moreover, the proposed completed local quartet patterns descriptor operator is a general texture descriptor which can be used in many other computer vision applications to describe image contents.