On-stream defect detection of metal artifacts from line CCD image

In this paper, we focus on a defect detection of a metal artifact, which has cylinder shape and hair-line finished surface, and it is sometimes ruined by scratch and/or liquid spill in factory line. So we have to find them from line CCD scanned images. In this paper, we employ Support Vector Machines (SVM) as classifier and test several transformations of input images to set input vector. We also discuss tradeoff between detection accuracy and computation time.