On close contour presentation and matching problems with biomedical image applications

by Tang, Yingjie.

Abstract (Summary)
Image contour extraction problem is a challenging problem in biomedical image applications. A model-based approach is proposed in this thesis as a solution to the contour extraction problem. Initially, an image contour extracted using a deformable contour method is categorized into one of a set of predefined models using a cross correlation detection method on wavelet descriptors. Wavelet descriptors of the curvature function extracted from the contour are chosen as the contour representation method so that the two-dimensional close contour data are transformed into one-dimensional descriptors. The wavelet descriptors contain positional information and have the multi-resolution representation capability. The image contour is matched against the selected contour model using the landmark information. Landmarks are peaks and valleys of the contour curvature wavelet descriptors. The highly mis-matched contour segments generated by landmarks are identified for further processing. The resulting contour is extracted when the highly mismatched segments are fine-tuned using a regionalized a posteriori probability contour model that is not a part of this thesis work. Experiment results on MRI brain and knee contour extraction problems are most encouraging.
Bibliographical Information:


School:University of Cincinnati

School Location:USA - Ohio

Source Type:Master's Thesis



Date of Publication:

© 2009 All Rights Reserved.