Inspection of metallic surfaces using Local Binary Patterns

In this paper we propose the use of a texture feature called Linear Binary Patterns to inspect steal surfaces. To assess the proposed method, we have build two different databases. The first one contains 996 color images of steel bars illuminated with black light, where the defects were highlighted using penetrating liquid. The second dataset is composed of 1141 gray-scale images of steel bars without highlight. Comprehensive experiments using three different classifiers show that the proposed feature set is able to detect 91.5% and 95.6% of the defects on the first and second databases, respectively.