Automatic visual inspection

In this thesis we introduce the subject of automatic visual inspection. While automation is not new in the process industries it is only just making its presence felt in the batch and discrete parts manufacturing industries. An integral process in discrete parts manufacturing is that of inspection which of necessity must be automated if complete automation is to be achieved. Here we are concerned with the replacement of the visual and interpretative processes of the human inspector by computer-based systems with image acquisition and image analysis capabilities. Research into automatic visual inspection requires the setting up of a laboratory in which the many varied aspects of the subject can be investigated. These include image acquisition, lighting, optics, mechanical handling, image processing and algorithm development. The digitisation and display of images is a vital prerequisite for such a laboratory as indeed is the provision of a comprehensive range of image processing functions. The thesis describes such a laboratory together with provisions for its hardware and software. It describes the research into two application areas of automatic visual inspection which have been investigated in the laboratory. The first of these is the shape analysis of essentially flat objects using a technique called the concavity tree description of shapes. This application focusses on those objects that can be inspected by inspecting their silhouettes so that binary (2 level) images are readily obtained. The second application is the inspection of glassware where considerable importance is attached to image acquisition. In this application the emphasis is on the rapid processing of the gray scale (multilevel) image to produce a binary image which can be analysed for the defects we are looking for.