Visual system for punched card reading in textile industry

Abstract

In this paper we present a novel design and implementation of a vision-based system for reading punched cards. The new system is designed to visually retrieve information stored in punched cards and store them in digital format. The system is dedicated to types of cards used in mechanical Jacquard looms widely spread in different textile industries. Image processing techniques like searching for edges using gradient, normalized cross correlation, and image registration are efficiently utilized to face several challenges. The existence of different types of disturbances makes gradient-based search for the pattern holes insufficient to obtain zero error ratio. As expected image registration can cope with such a problem after efficiently solving the feature extraction and matching problem resulted from the fact that the card image does not contain any kind of information except the pattern holes which are not discriminating features. Experiments have been carried out using the built real system. Experiments show results with 0.041% error ratio. This ratio is highly satisfying the textile industry.