Ph.D. Thesis Summary
An Analytical and Experimental Investigation of Physically-Accurate Synthetic Images for Machine Vision Design
Johné Michelle Parker
For machine vision system applications such as part presentation, the accuracy of image gray-scale pixel values far outweighs image appearance; this thesis presents physically-accurate image synthesis as a rational basis for designing both hardware and software components of a vision system. This is a complex task, since such systems consist of many parts; the most proficient systems are designed by considering the integrated hardware/software arrangement. Numerical simulation (i.e., physically-accurate image synthesis) is a flexible, practical tool for investigating a large number of hardware/software configuration combinations for a wide range of parts.
While synthetic images presented in this work do not fully emulate the real manufacturing environment, they can be efficiently used to 1) predict the captured-image intensity values produced by a specific machine vision system and 2) study the effects of major design parameters, such as ambient lighting, illumination distribution, true part and environment reflectance and geometric alignment on image accuracy. Additionally, this study shows that physically-accurate simulations of image capture can differentiate between minute changes in design factors, facilitating an improved understanding of the effects that variations in these intrinsic parameters have on image generation and processing.
Major research benefits are three-fold: Physically-accurate image
synthesis enables a reliable prediction of a design's performance characteristics
prior to actual implementation. Secondly, this research introduces
a methodology for comparing algorithms and predicting the optimal algorithm
(and optimal performance) for specific applications. An in-depth
study of the factors that significantly degrade the performance of image-processing
algorithms indicates critical design parameters. A third contribution
is that this investigation provides a solid foundation for the development
of a CAD-tool which utilizes physically-accurate synthetic images to accurately
and inexpensively predict the performance of a proposed vision system design
prior to implementation or the construction of a prototype.