Identification of sticker printing defects in glove manufacturing process using Computer Vision techniques

Abstract

This research aims to provide an automatic and real-time defect detection framework for the glove manufacturing industry using computer vision techniques. This research concerns detecting defects on specific sticker printed on the glove at the end of the production. The whole approach is divided into two operational modes: Teaching mode and Inspection mode. The teaching mode contains time complex tasks that can be performed before the actual inspection. The inspection mode does the actual inspection to find the defects. An image of a printed sticker will be processed in inspection mode using three levels to identify defects. Lower levels contain naïve computer vision algorithms and detect high-degree errors only, whereas higher levels contain complex algorithms that could detect more sophisticated errors. It is an efficient technique to identify defects in the early stages of the defect inspection process. The significance of sticker's content to its domain will be calculated for every object in the sticker by combining the visibility and domain importance of that specific content. The visibility of content is measured using size and density. A decision function is proposed to decide whether to accept or reject the glove by considering the calculated error and the significance. Finally, a quality measurement model is proposed to calculate the printed sticker's quality for each accepting glove. The visibility calculation model proved to be valid and consistent with perceptual visibility. The significance calculation model also provides reliable and consistent results according to testing. The defect inspection process is also efficient and performs as expected. However, inspection level-3 provide inconsistent result in some situations, and that algorithm needs to be improved.