Real-time Industrial Vision System for Automatic Product Surface Inspection

Abstract

Product surface inspection plays a significant role in industrial aspects. Large industrial manufacturing requires such inspection procedure of high speed and accuracy at a fairly reasonable cost, which is precisely the demand automatic surface inspection systems are applied to meet. In this paper, we have constructed a vision system prototype employing image processing and pattern recognition approaches to classify those defective products automatically. Our algorithm first collects products images, then send them to preprocess. After that, we implement pattern extraction based on Fourier-Mellin transform, and classify the product patterns based on principle component analysis as well as support vector regression. The prototype has proven itself reliable through reaching accuracy of more than 90%.