Application of local binary patterns to surface defect recognition of continuous casting slabs

To solve the detection problems of slab surface defects by conventional image recognition algorithms, this article introduces an improved multi-block local binary pattern algorithm which considers the image's pixels. In this algorithm, the original image is divided into several small regions, each small region is equally divided, and the average gray value is calculated. Then the local binary pattern algorithm is used. Five different kinds of 1697 samples gathered from a production line of slabs were examined, including cracks, scratches, indentations, dents, and no defect. The recognition rate reaches 94.9%, while the recognition rate of the traditional local binary pattern method is 89.1%. The results show that the proposed algorithm has the characteristics of high precision, better robustness and noise immunity.