Industrial machine vision, using cameras and computers to automate industrial inspection tasks, has proven to be harder to introduce into factories than was anticipated. The major reason is that the underlying principles for extracting industrially useful information from images are not well understood. The objective of this research is to construct a new research tool for studying the information content in industrial machine vision images: a computer simulation of image capture systems. This simulation will combine numerical models of lighting, surface reflection, optics, cameras, and video electronics to synthesize images with highly accurate surface luminances. Using the simulator to replace a vision system's camera, image processing algorithms can be tested by applying them to images with known information and noise characteristics. When these images are compared with the outputs of cameras, insight into the nature of camera defects will also result. In short, this simulator provides the first tool that can look into all areas of an image capture system, even those that are normally inaccessible. The principal investigator expects this to lead to new ideas about the design of image capture systems and image processing algorithms. In addition, the simulator has a practical use. Because it can simulate images of parts that have not yet been built, it can be used as a tool to design industrial vision applications or as part of a system that automatically programs vision systems.
