Electronic Theses and Dissertations



Document Type


Degree Name

Doctor of Philosophy


Electrical & Computer Engineering

Committee Chair

Madhusudhanan Balasubramanian

Committee Member

Deepak DV Venugopal

Committee Member

Nirman NK Kumar

Committee Member

Aaron AR Robinson


The 3D architecture of an object or of a scene can be estimated non-invasively from estimates of dense geometrical correspondences among two (stereo) or more views (2D projections) of the scene. Establishing dense correspondences among geometrical coordinates of multiple views (lower-dimensional projections) is an ill-posed problem due to scene occlusion. In addition, larger scene extents with smoother texture characteristics and differences in scene illumination among the multiple views/observers are sources of difficulties in disparity estimation. A successful strategy for improving the accuracy of the disparity estimates is to utilize spatial dependencies of the scene characteristics as well as of the disparity estimates. Among the probabilistic inference formulations for estimating dense geometric correspondences, \textit{Markov Random Fields} (MRF) based approaches have been successful in modeling spatial geometrical dependencies. One of the limitations of the MRF models is that the \textit{neighborhood system} or \textit{clique} used for enforcing spatial dependencies is required to be \textit{maximal}. Further, the chosen MRF dependency structure is uniformly enforced for all the random variables on the pixel lattice. While learning methods are available for optimizing both the MRF parameters and the MRF neighborhood structure, they are generally limited to specific tasks. Therefore, lower-order MRF models are generally used for various computer vision problems including disparity estimation. In this dissertation, we present \textbf {1.} a computationally efficient \textit{maximum likelihood} strategy to estimate stereo disparity from sparse disparity cost volumes (HCS algorithm); \textbf{2.} a new factor graph-based probabilistic graphical model (FGS algorithm) for disparity estimation that addresses the aforementioned MRF limitations by allowing a larger and a spatially variable neighborhood structure determined based on the local scene characteristics; \textbf{3.} a new multi-resolution factor graph-based disparity estimation framework (MR-FGS algorithm) that provides higher accuracy and sharper disparity boundaries than the FGS algorithm; \textbf{4.} comparative evaluation of 3D geometries of scenes estimated using our new disparity estimation algorithms. We evaluated the proposed algorithms using the \textit{Middlebury benchmark stereo datasets} and the \textit{Middlebury evaluation dataset version 3.0} and compared their performance with recent state-of-the-art disparity estimation algorithms. Our factor graph formulation can be useful for obtaining \textit{maximum a posteriori} solutions to optimization problems with complex and variable dependency structures as well as for other dense estimations problems such as optical flow estimation.


Data is provided by the student.”

Library Comment

Dissertation or thesis originally submitted to ProQuest.


Open Access