Ph.D. Dissertation Defense “Recognition and Synthesis of Incomplete Objects Using a Geometric Based Local-Global Graph Description” By Michael Christopher Robbeloth

Ph.D. Committee:  Drs. Nikolaos Bourbakis, Advisor, Soon Chung, Yong Pei, and Arnab Shaw (EE)

ABSTRACT

The recognition of single objects is an old research field with many techniques and robust results. The probabilistic recognition of incomplete objects, however, remains an active field with challenging issues associated to shadows, illumination and other visual characteristics. This dissertation presents a suite of high-level, model-based computer vision techniques encompassing both geometric and machine learning approaches to generate probabilistic matches of objects with varying degrees and forms of non-deformed incompleteness. 

The recognition of incomplete objects requires the formulation of a database of six-sided exemplar images from which an identification can be made. The images are broken down by different algorithms followed by region and segment isolation along with object level rotation and segment level synthesis from which geometric and characteristic properties are generated in a process known as the Local-Global (L-G) Graph method. The properties are then stored into a database for processing against sample images featuring various missing features or non-deformed distortions in a multithreaded manner. The multi-factor matching procedure uses weighted measures to determine the likelihood of a sample image matching one of the exemplars. The ability to find a match is extensible in the future by adding additional detection methods.

Overall, the results, while promising, show that there is still much work to be done. It is evident that there are many additional avenues to explore related to different detection methodologies along with performance enhancements to be employed across both computational and memory constrained resources to drive the recognition of incomplete objects in production systems.