Patterned Fabric Inspection and Visualization by the Method of Image Decomposition

Abstract

This paper analyzes the cartoon and texture structures to inspect and visualize defective objects in a patterned fabric image. It presents a method of an image decomposition (ID) and solves it by a convex optimization algorithm. Our experimental results on benchmark fabric images are superior to those by other methods.Note to Practitioners-This paper is motivated by an ID method to examine how to novelly represent defective objects and repeated patterns in fabric images. We decompose a fabric image into two components: cartoon structure as defective objects and texture structure as repeated patterns. The ID is optimized by the largest correlation between a given defect-free fabric image and the texture structure of a testing image. Its merit is requiring only one defect-free image to optimize the inspection. The resulting cartoon structure is identified for inspection and visualization. An intensive performance evaluation is conducted on dot-, star-, and box-patterned fabric images and the detection accuracies range from 94.9% 99.6%. This research is beneficial to the practitioners for quality control in textile, ceramics, tile, wallpaper, printed circuit board, and aircraft window industries.

Combining Fisher Criterion and Deep Learning for Patterned Fabric Defect Inspection

“…To evaluate the performance of FCSAE method, we compared it with ID method [2] and SAE method. The boxpatterned fabric images used in [2] were employed as the benchmark images.…”

“…The boxpatterned fabric images used in [2] were employed as the benchmark images. There are five types of defects in the dataset, namely, "Broke End", "Hole", "Netting Multiple", "Thick Bar" and "Thin Bar", and each type has 5 pictures.…”

“…2, we can see that the results of FCSAE are more accurate. Several measurement metrics such as ACC, TPR, FPR, PPV and NPV [2] were further employed to quantify the detection accuracy. Here we adopt the average of these metrics to demonstrate the detection accuracy.…”

“…Compared with detecting plain and twill defects, patterned fabric inspection is far more complex. A summary of methods of patterned fabric defect detection can be found in earlier work [1], [2]. Ng et al presented a novel method of decomposing the fabric image into a cartoon structure and repeated patterns according to the image decomposition method (ID), which is superior to other methods on benchmark images [2], [3].…”

Patterned fabric defect detection via convolutional matching pursuit dual-dictionary

“…As a result, the key feature of the fabric images becomes blurred and the intrinsic information of image is inundated, 21 which not only increases the difficulty of image analysis but also affects the final test results of defect detection. As a result, the key feature of the fabric images becomes blurred and the intrinsic information of image is inundated, 21 which not only increases the difficulty of image analysis but also affects the final test results of defect detection.…”