Blind weave detection for woven fabrics

For the purpose of textile quality assurance, an algorithmic framework for fully automatic detection of weave patterns in woven fabrics is presented. The proposed method is able to handle fabrics of any rotation, material, and binding. Periodicity features within highly resolved fabric images are found and structured in a compact yarn matrix representation which allows to estimate the trajectories of single yarns. Fourier analysis, template matching, and fuzzy clustering are some of the key methods employed during the process. From the yarn matrix, the fabric's weave and density can directly be derived. Since a multitude of factors may falsify the output, a feedback loop is integrated to iteratively find an optimal result. The framework works completely blind, i.e., without any a priori knowledge of the fabric. The evaluation has been conducted on an extensive image database of 140 real-world fabric images including cotton, polyester, viscose, and carbon materials of plain, twill, or satin weave. The system proved to be robust and versatile as a 97 % detection accuracy could be achieved. Source codes and image databases are provided.