An approach to optimal and dense stereo image correspondence

• Main Author: Galarza, Luis E.
• Format: Others
• Published: FIU Digital Commons 2005
• Subjects: Engineering
• Online Access: http://digitalcommons.fiu.edu/etd/3607

Stereo correspondence is used in the extraction of the three dimensional structure of a scene. This is possible because the difference or displacement in position of corresponding Image points (also known as disparity) is related to the three dimensional position of the object point. In the stereo correspondence problem, the object is to determine the locations in each image that are the projection of the same physical point in space. No general solution to the correspondence problem exists mainly due to ambiguities into what constitutes a match. To resolve these ambiguities constraints and assumptions must be made. This thesis approaches this problem using an optimal and dense image matching procedure. The procedure is based on a method where epipolar and ordering constraints are employed. New to this approach is that this inherently constrained problem is solved in an unconstrained manner. In addition, reliance on heuristics is significantly reduced. The thesis presents a one dimensional and a two dimensional approach for addressing this problem. The two dimensional approach is a generalization of the one dimensional method such that spatial correlations inherent in images can aid in the matching process of inherently noisy images.