Fabric defect inspection based on lattice segmentation and template statistics

Automated fabric inspection is desirable for quality control of fabric industry. However, it is challenging due to some unpredictable fabric defects which may only occur during the production. Hence, methods aiming at automated fabric inspection are commonly developed in absence of defective samples. This paper proposes a novel automated fabric inspection method based on lattice segmentation and template statistics (LSTS) focusing on the patterned fabric images containing repetitive texture primitives. The proposed method attempts to infer the placement rule of texture primitives and divide the image into none-overlapping lattices as texture primitives which represent the given image by hundreds of lattices instead of millions of pixels. The defects are then efficiently identified by comparing the lattice similarity w. r. t. the benchmarks named template statistics. The template statistics are learnt from defect-free samples through a modular framework in which multiple feature extraction methods like Gabor filters and local binary pattern can be flexibly combined according to their inspection efficiencies. The performance of the proposed method is evaluated in the databases of dot-, box- and star-patterned fabric images. By comparing the resultant and ground-truth images, an overall detection rate of 0.977 is achieved, which is competitive with the state-of-the-arts. (C) 2019 Elsevier Inc. All rights reserved.