Label Inspection using the Hough Transform on

A variation on the Hough transform, the special normal/angle Hough transform , (RsHT) is described detailing its characteristics and advantages and suitablity for a parallel implementation. The transputer is introduced and the application of the RsHT on two popular types of transputer networks, a control-driven array and a demand-driven farm of transputers is described. The methods employed for communication and the breakdown of the algorithms are described for each type of the network. Timing results for the implementation on both networks are provided. The use of the RsHT is then further examined with application to the problem of label inspection and two types of labels are used to demonstrate how to pass or fail labels under inspection. 1 Introduction The detection and veriication of labels stuck on products is generally termed label inspection , and is often the nal task in the manufacturing process. The speed of manufacturing generally forces the acquisition of dedicated hardware for the inspection. However, this restricts the exibility of the inspection algorithms that can be used. This restriction can be eliminated by use of parallel processing architecture and software, such as an expandable transputer connguration. In this study, a variation of the Hough transform has been employed to provide data which, after further processing, may be used to make a pass or fail decision on the item under inspection. The Hough Transform is a com-putationally expensive task, but is well suited to a parallel implementation. Therefore, the performance when implemented on a networks of parallel processors using diierent programming models is considered. In the next section, the Hough Transform and its variant, the special normal/angle Hough transform (RsHT), are described, showing why the RsHT is suitable for parallel processing. Section 3 describes the transputer. Section 4 covers the programming models, and types of transputer networks used, and the implementation of the RsHT on the respective networks. Section 5 expands on the problem of label inspection and presents some implementation results. In conclusion, section 6 shows how the RsHT results may be used to detect faulty labels. 2 The Hough Transform Formulation The straight line Hough transform formulation used in this application is the normal/angle transform originally proposed by 1]. This is an improved formulation over that proposed