Enhanced RGB Image Processing Method for Automatic 2D and 3D Defect Inspection on Shiny Surfaces

Defect inspection using computer vision is one of the most widespread artificial intelligence application. However, this technique has some serious limitations. Images from high reflective flat surfaces such as metal sheets may contain high amounts of glare, which makes them unsuitable for training, as defects may be hidden beneath disturbing reflections. Moreover, 3D defects may not be detected with conventional 2D camera. Finally, AI systems rely on large labelled image datasets for training, which may be extremely difficult and expensive to obtain. To solve these issues, in this paper we propose a procedure for detecting 2D and 3D defects on metal blanks based on an RGB camera, a laser light source and a limited defect dataset.