Orasis Introduction

Orasis project goal is to apply Spectral Graph Theory to the image processing problems. The first application, and possibly the most pressing issue, in image processing is the segmentation problem. Historically most segmentation methods determine the segmentation based on fairly local consideration. I classic example are the level set methods. In this project we have developed methods that use the entire picture to segment the image. Our new method is called Spectral Rounding (SR). SR has been should to give high quality segmentation even for very noisy data.

SR uses eigenvectors as well as the solution to symmetrical diagonally dominate linear systems. Until recently computing these needed eigenvectors and solutions has been very time consuming. Fortunately we have also developed new solvers of 2D images that work in linear time in the size of the image.

Gary Miller

Last modified: Fri May 25 14:52:43 EDT 2007