Defect Detection in Textiles with Co-occurrence Matrix as a Texture Model Description

Abstract

Automatized inspection at textile production lines becomes very important. However, there is still a need to design methods which meet not only demands concerning accuracy of defect detection, but also ones related to the processing time. In this work, a novel approach for defect model definition is presented. It is derived from the idea of co-occurrence matrix. Due to scale incorporation and binarization of the model content it proved to be a very powerful descriptor of the novelties. Moreover, it also satisfies the requirements of short processing time. The defect mask achieved with the introduced method was compared visually to other popular solutions and show a very high accuracy and quality of defect description. The processing time is real-time as the response for a 1MP (megapixel) image is reached within tens of milliseconds.