Masters Theses

Date of Award

12-1989

Degree Type

Thesis

Degree Name

Master of Science

Major

Electrical Engineering

Major Professor

Dragana Brzakovic

Committee Members

Molan M. Trivedi, M. M. Abidi

Abstract

An iterative image segmentation algorithm utilizing image pyramids with fuzzy links is presented. The concept of a membership function is used to establish links between nodes on adjacent levels of the pyramid. The base of the pyramid is the image to be segmented. Each subsequent level of the pyramid is derived by convolving the preceding level of the pyramid with a Gaussian mask in an overlapped manner and has half the spatial resolution of its predecessor. Nodes at adjacent levels of the pyramid are linked in a bottom-up manner using a membership function which is chosen to yield the desired segmentation result. The pyramid is recomputed based on the previous links and a new set of links is defined. This iterative process culminates when the links reach a steady state. Segmentation is achieved in one top down pass, whereby nodes on a higher resolution level are replaced by nodes from the immediately adjacent lower resolution level above based on the strength of the links The fuzzy pyramid linking algorithm is a generalization of the classical 'hard' pyramid linking algorithm, and can be tuned to produce the desired segmentation results. While the fuzzy pyramid linking procedure is computationally more demanding, it greatly improves the quality of the segmentation and is guaranteed to converge in a finite number of iterations.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS