K.U.Leuven > ESAT > PSI > Visics > Research > Topics > Item 3.1

Textile inspection

L. Van Gool, E. Pauwels, J. Van Rompay

In this IWT project, carried out together with the company Barco NV (Barco Automation) and the Rijksuniversiteit Gent, methods were developed for the detection and classification of flaws in woven textiles, in particular denim. The method is based on the design of a set of dedicated filters, the shape and parameters of which are learned automatically from a flawless example textile. The filters are the results of a principal component analysis of the intensity variations within one period of the fabric. They are kept sparse, i.e. have many zero elements, in order to ensure fast processing. These convolution filters are then applied to images of the textile, on-loom. Abnormal values of the filters' output highlight the positions of the flaws. A more precise analysis of the values of the different filters and the overall shape of the flawed region gives further information about the specific types of error.

The left image shows the input image. The right image shows the highlighting of flaws after filtering techniques.