Automated real-time visual inspection system for high-resolution superimposed printings

https://doi.org/10.1016/S0262-8856(98)00059-6Get rights and content

Abstract

This paper describes an automatic real-time inspection system of printed products. This system has been applied to the inspection of objects having superimposed printings with different reference systems at high resolution, that is, there are not two identical products. The automation of this process is particularly critical and its solution is highly complex. The techniques employed are based on a combination of various disciplines, allowing a compromise between resolution and processing speed. In this way, disciplines such as morphological processing, subpixel accuracy, scheduling techniques, parallel processing, etc. are incorporated.

Introduction

The development of an automated visual inspection system for high resolution superimposed printings requires the use and refinement of various techniques.

The large volume of data to be handled by the application, along with the demanding requirements on the production line, demand the use of algorithms for parallel processing of images. Attempts to minimize processing time require the use of different optimization strategies depending on the level at which work is being carried out.

In some lower-level processes such as point-to-point operations between images, the total image is partitioned in such a way that it is distributed according to regions dependent on each Image Acquisition and Processing Board (IAPB). Thus, each one of these boards will act in parallel with the others.

Other low-level processes, such as transfer of image areas, require the combined use of the CPU and some Image Acquisition and Processing Boards. In such cases, an optimization of the system resources is necessary. Furthermore, high-level processes must be executed by the system CPU, coexisting with the other processes as described above.

The foregoing paragraphs show the need to carry out proper time and space distribution of tasks to be handled by the existing processors, in order to attain an optimal performance of the application. It should be pointed out that this scheduling will depend on the features of the product to be inspected. Therefore, it should be flexible, since the scheduling may have to be modified depending on the application.

However, the implementation of an automated visual inspection system for high resolution superimposed printings in real time requires more than just the techniques described in the preceding paragraphs. In the field of machine vision inspection, we can also find the difficulties associated with the template-matching technique. This paper describes a solution implemented through a new philosophy in the configuration process of pattern images based on morphological operators with subpixel precision.

To summarize, this paper presents the conclusions drawn in different scientific and technological fields which, when brought together in a coherent manner, offer an automated visual inspection system for products with various superimposed high-resolution printings. The system has been applied to the quality control of bills, a product which has several superimposed multilevel printings.

Section snippets

System structure

The system is made up of two parts. The first is a mechanical module designed to position the sheets of bills to be inspected correctly in order to aid in the scanning of images. The other part consists of a computer vision module supported by hardware and software elements, as outlined in Fig. 1.

The vision system consists of the following hardware elements: a control module, an image processing module and an acquisition module.

Distribution of images among processors

By using parallel processing techniques it is possible to achieve improvements in two areas; processing time and the quality of the process to be inspected. Assuming that parallel processing is necessary due to the high volume of data to be processed in real-time, one has to establish the amount and type of data corresponding to each stage of processing.

Nevertheless, there are a series of determining factors that also affects in ascertaining the degree of parallel processing to be used in a

Scheduling issues

Computer vision applications, which must offer a response in a set period of time, are real-time applications in their broadest sense. As in other real-time systems, in computer vision applications used in the field of industrial inspection, the requirements of a temporal nature are generally known and fixed quantities. In other words, these systems are driven by time.

In order for this idea to materialize, there must exist some type of communication and synchronization among the components that

Inspection algorithm

The most commonly used inspection algorithms, both in binary and grey-level images, use different methods based on techniques such as: comparison with reference or pattern images, matching 6, 7, 8; rule testing by checking dimensional characteristics 9, 10, 11; morphological transformations applied to the pattern image, which result in abstract figures in which only information of interest remains 12, 13; differential scanning with no need for pattern images [6]and spatial filtering using

Application to the inspection of superimposed images

In automatic inspection via computer vision of printed products, the complexity increases as new variables are introduced: multilevel printing, superposition of them, each printing with a different system of reference, necessary speed, available time for carrying out the inspection, etc.

All of these factors require the use of the different techniques explained above (scheduling, morphological processes, detection with subpixel precision, etc.) in order to achieve a robust and reliable system

Application to the inspection of bill-sheets

In the automatic inspection of bill-sheets using computer vision, the greatest degree of difficulty is encountered when several multilevel printings are superimposed. This problem has been tackled by R&D departments of mints all over the world.

Conclusions

One of the most noteworthy characteristics of the systems is its applicability to various disciplines.

It is also worth mentioning the correlation sought between real-time systems and computer vision using a parallel architecture which grants all the necessary requirements both in the area of time and in the resolution required for the inspection are of noteworthy mention.

In reference to the inspection methodology, the applications developed in the area of pattern image generation using

References (19)

  • P.E. Danielsson et al.

    Distance checking algorithms

    Computer Graphics Image Process.

    (1979)
  • T.S. Newman et al.

    A survey of automated visual inspection

    Comput. Vision Image Understand.

    (1995)
  • Matrox, The Image-Series General Introduction Manual, Matrox Electronic Systems Limited, Quebec, Canada,...
  • D.I. Moldovan, Parallel Processing from Applications to Systems, Morgan Kaufmann, San Matco, CA,...
  • P.P. Jonker, Morphological Image Processing: Architectures and VLSI Design, Kluwer, Netherlands,...
  • E. Embrechts, D. Roose, P. Wanbacq, Component Labelling on a MIMD Multiprocessor, Cvgip:Image Understanding, Academic...
  • F. Torres, Parallel architecture for the high-resolution image processing. PhD thesis, Polytechnic University of...
  • R.T. Chin

    Automated visual inspection

    CVGIP

    (1988)
  • J. Weng

    A theory of image matching

    Int. J. Comput. Vision

    (1990)
There are more references available in the full text version of this article.

Cited by (0)

View full text