Intelligent Machine Vision

Vision is critical for the survival of almost all species of the higher animals, including fish, amphibians, reptiles, birds, and mammals. In addition, many lower animal phyla, including insects, arachnids, Crustacea, and molluscs, possess well-developed optical sensors for locating food, shelter, a mate, or a potential predator. Even some unicellular organisms are photosensitive, even though they do not have organs that can form high-resolution images. Vision bestows great advantages on an organism. Looking for food, rather than chasing it blindly, is very efficient in terms of the energy expended. Animals that look into a crevice in a rock to check that it is safe to go in are more likely to survive than those that do not do so. Animals that use vision to identify a suitable mate are more likely to find a fit, healthy one than those that ignore the appearance of a possible partner. Animals that can see a predator approaching are more likely to be able to escape capture than those that cannot. Compared with other sensing methods based on smell, sound, and vibration, vision offers far greater sensitivity and resolution for discriminating the four essentials for survival listed above.

The purpose of this chapter is to outline some of the basic techniques used in the development of industrial machine vision systems. These are discussed in sufficient detail to allow an understanding of the key ideas outlined elsewhere in this book. In the following discussion we will frequently indicate the equivalent PIP operators for the vision techniques described. PIP commands appear in square brackets. In certain cases, sequences of PIP commands are needed to perform an operation and these are similarly listed.

This chapter, indeed the whole of this book, is based on three axioms:

The primary purpose of this chapter is to discuss image-processing implementations on desktop computers (PCs) and special-purpose hardware. The preliminary discussions apply to both kinds of systems. Although most readers will be more familiar with the former, they are asked to keep both possibilities in mind.

Writing code to implement just one of the simpler image-processing operators, such as negate(neg/0), threshold (thr/2), etc., presents no difficulties whatsoever, even to a novice programmer. However there is a large difference between writing code to implement a single isolated function and constructing a reliable software package that has a comprehensive command repertoire and a well-engineered user interface. In this chapter, we will explore some of the general issues that a programmer should bear in mind when implementing an interactive vision system for use in Machine Vision. We will not discuss the software implementation of specific imageprocessing functions at great length, since several excellent books containing C and Java programs for image processing, already exist [WHEOO]. We will review the overall structure of several IVS packages, including:

PIP and WIP are two of the latest developments in interactive vision systems that began In the mid-1970s, with SUSIE. All of the systems built by the authors since then operated In stand-alone mode. There have been brief and tentative incursions into networking, although these have never been fully developed, because at the time there was no universally accepted protocol for worldwide digital communication [BAT91a]. The advent of Internet technology¹ has made it possible to develop new avenues of approach that were not possible, nor even anticipated, in the mid-1990s. The Internet revolution is just beginning and already owes a great deal to the Java programming language, which was the chosen vehicle for developing CIP, the immediate successor to PIP. However, we will see in this chapter that our plans for a Web-based toolbox for vision system designers is much more ambitious than this.

This chapter will outline the development of a visual programming environment for machine vision applications, namely JVision ² [WHE97a, WHE97b]. The purpose of JVision is to provide machine vision developers with access to a non-platform-specific software development environment. This requirement was realized through the use of Java, a platform-independent programming language. The software development environment provides an intuitive interface which is achieved using a drag-and-drop block-diagram approach, where each image-processing operation is represented by a graphical block with inputs and outputs which can be interconnected, edited, and deleted as required. Java provides accessibility, thereby reducing the workload and increasing the “deliverables” in terms of cross-platform compatibility and increased user base. JVision is just one example of such a visual programming development environment for machine vision. Other notable examples include Khoros[KR199] and WiT [W1T99]. See [JAW96, GOS96] for details on the Java programming language.

Machine Vision has been studied in universities, commercial companies, and various other research organizations for over 20 years. The world-wide market for vision-based products and services for industrial, medical, security, and other applications is already about US$4000 million per annum and is growing rapidly. However, it must be admitted that industrial Machine Vision is still in its infancy, in the sense that the technology is not widely understood by the general engineering and business fraternity. As a result, it is often mistrusted, misused, and underused. Recall that in Chapter 1 we pointed out that most people, including technologists, erroneously believe that they can explain how human beings see. It has been found by experience that “obvious” solutions to Machine Vision applications almost invariably do not work. Indeed, even an experienced vision engineer can only rarely predict correctly what algorithm will be most appropriate, if he is limited to viewing an object or scene by eye. The applications studies described in this chapter are intended to demonstrate this point, plus two others:

When the authors began working (separately) In Machine Vision research in the mid-1970s, the subject was still very much in its infancy. The limited computing power then available made it very difficult to apply image processing to any but the simplest and least demanding inspection tasks. Since then, the subject of Machine Vision has been transformed by several key innovations, which have greatly improved: