A comparative study of image processing thresholding algorithms on residual oxide scale detection in stainless steel production lines

Abstract

The present work is intended for residual oxide scale detection and classification through the application of image processing techniques. This is a defect that can remain in the surface of stainless steel coils after an incomplete pickling process in a production line. From a previous detailed study over reflectance of residual oxide defect, we present a comparative study of algorithms for image segmentation based on thresholding methods. In particular, two computational models based on multi-linear regression and neural networks will be proposed. A system based on conventional area camera with a special lighting was installed and fully integrated in an annealing and pickling line for model testing purposes. Finally, model approaches will be compared and evaluated their performance..