Automated visual inspection of textile

A method for automated inspection of two types of textile is presented. The goal of the inspection is to determine defects in the textile. A prototype is constructed for simulating the textile production line. At the prototype the images of the textile are acquired by a high speed line scan camera. The vertical threads are located using a vertical projection of the image.

It is thereby possible to identify the defects in the vertical threads. A structural model of the horizontal threads is formulated. The model consists of a Markov random field which represents a priori knowledge about the position and structure of the horizontal threads and an observation model that incorporates knowledge about the visual appearance of the threads given their position and structure.

Using this model the horizontal threads are located. Features are calculated from the located threads to identify the defects. To go from the protype to a production line system we only need to gain a speed factor of 4.