Advancing data discovery through new elliptic and related higher-order shape-based processing
Timothy S. Newman, Chunguang Cao · 2007
This dissertation investigates using elliptic shape and related higher order shapes in image segmentation and object detection. Three new methods for ellipse detection, ellipsoid detection, and deformable ellipsoid localization are introduced. One primary focus of this dissertation is a new ellipse detection method. The new method is based on least-squares fitting and randomized Hough transform (RHT). It overcomes some problems of the existing ellipse detection methods. The new method is demonstrated in its primary application in segmentation of auroral ovals from satellite imagery, as well as in the content-based image retrieval of auroral images from an image archive. An extension of the ellipse detection method to ellipsoid detection is also presented. The ellipsoid detection method utilizes geometric properties of the ellipsoid for centroid evidence accumulation through RHT. The axial and orientation information of the ellipsoid is recovered using least-squares fitting-based RHT and principal component analysis. Experiments on both synthetic and real data show that the ellipsoid detection method is efficient and accurate. The third new method introduced in this dissertation involves a higher order form of the ellipse, namely, deformable ellipsoids. The method allows deformable ellipsoids to be localized. A demonstration of using the deformable ellipsoid for right ventricle localization in a type of volumetric medical imaging data is presented.