Optical Crack Detection of Refractory Bricks

In this work we show an approach to detect cracks in refractory bricks. The bricks have to be inspected while they are moving through the production line by chain conveyers. The cracks which have a width of down to 100 micrometer are very small compared to the dimension of the bricks. The surface of the bricks can vary in colour, lightness and roughness even if the bricks are part of the same batch. A non-contact optical measuring system is able to meet these requirements. The approach we present in this paper works on images we receive from an industrial CCD-camera. One of the main difficulties in analysing these images is that we have to distinguish between real crack patterns and normal roughness of the brick-texture. The presented algorithm is split into four steps. The first includes image pre-processing to reduce resolution (for faster computation) without losing important information on the thin cracks, background subtraction and denoising. At the second step we use Gabor filter in different directions to enhance crack-like structures and get information about direction. Next a tracing-algorithm, which starts at thresholded seedpoints, let the cracks grow. At the final step, the possible cracks detected by the tracing algorithm, are checked for their length. If a crack obtains a certain length the brick needs to be rejected.