Automated assessment of textile seam quality based on surface roughness estimation

Abstract

In this paper the issue of automated seam quality control is addressed, focusing especially on seam pucker evaluation. Currently this task is accomplished by human experts considering five grades of quality. The proposed method estimates surface roughness of seam specimens producing robust and efficient novel features highly correlated to quality grades (QGs). At the initial stage, oblique illumination is applied and two-dimensional images of the specimens are acquired. The images are automatically rotated and centered in respect to the seam line and segmented into four regions. Each region produces an intensity curve through averaging, and roughness estimation is performed based on intensity mean deviation. Finally, a QG is assigned to each specimen using a k-nearest neighbor classifier (kNNc). A data set containing 211 seam specimens, created by two different kinds of fabric, has been used for testing and a correct classification rate of 81.04% has been produced matching up to the performance of human experts.

Keywords:

• machine vision
• seam pucker
• quality control
• kNNc