Machine learning algorithms have been successfully utilized in various systems/devices. They have the ability to improve the usability/quality of such systems in terms of intelligent user interface, fast performance, and more importantly, high accuracy. In this research, machine learning techniques are used in the field of image understanding, which is a common research area between image analysis and computer vision, to involve higher processing level of a target image to "make sense" of the scene captured in it. A general probabilistic framework for image understanding where topics associated with (i) collection of images to generate a comprehensive and valid database, (ii) generation of an unbiased ground-truth for the aforesaid database, (iii) selection of classification features and elimination of the redundant ones, and (iv) usage of such information to test a new sample set, are discussed. Two research projects have been developed as examples of the general image understanding framework; identification of region(s) of interest, and image segmentation evaluation. These techniques, in addition to others, are combined in an object-oriented rendering system for printing applications. The discussion included in this doctoral dissertation explores the means for developing such a system from an image understanding/ processing aspect. It is worth noticing that this work does not aim to develop a printing system. It is only proposed to add some essential features for current printing pipelines to achieve better visual quality while printing images/photos. Hence, we assume that image regions have been successfully extracted from the printed document. These images are used as input to the proposed object-oriented rendering algorithm where methodologies for color image segmentation, region-of-interest identification and semantic features extraction are employed. Probabilistic approaches based on Bayesian statistics have been utilized to develop the proposed image understanding techniques.
Library of Congress Subject Headings
Machine learning; Image processing--Digital techniques; Bayesian statistical decision theory; Computer vision
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Jaber, Mustafa, "Probabilistic framework for image understanding applications using Bayesian Networks" (2011). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus