| This dissertation presents a tractable and empirically accurate algorithm, along with an integrated framework, realizing a mid-level visual process for pattern recognition. The algorithm and overall framework takes advantage of hypotheses provided by a high-level visual process, thereby, attempting to extract a region in an image based on these hypotheses. The main focus is to recognize analytical, as well as non-analytical objects from gray-level images. The overall approach is based on a study of the Hough Transform and its generalized version.; A complete discussion on the current state-of-the-art covers several different methodologies used to solve this problem. Each of these techniques are reviewed and highlighted.; An examination of the Hough Transform and review behind the underlying theory for the generalized version is presented. Furthermore, a discussion on additional low-level visual process concerns, related to the dissertation's approach and otherwise, is reviewed. The low level visual process is an important part of the overall framework, the theories as well as practical implementations are examined.; A novel algorithm based on the generalized Hough Transform, and an integrated framework of support processes, is subsequently presented. The new approach exploits the geometry involved with parallel gradient angles around an object. As a result, a new version of the generalized Hough Transform's R-Table is developed which is used to effectively provide hypotheses to the mid-level visual process. The new algorithm, associated with its inverse operation and additional procedures, allows the extraction of desired regions within images.; To show the overall usefulness of the framework, an extensive series of experiments are performed. These experiments test the novel algorithm's capabilities under several conditions, as well as the overall framework itself. An analysis of absolute and percent error accuracy is presented, indicating the effectiveness of this dissertation's approach.; Finally, the document concludes by summarizing the underlying theory, experiments performed, and inferences obtained. The work closes by considering future areas of study opened up through this research, as well as opportunities for practical use. |
