HR: 0800h
AN: SM51B-1285    [Abstracts]
TI: A Shape-Based Technique for Aurora Oval Segmentation From UVI Images
AU: * Cao, C
EM: ccao@cs.uah.edu
AF: Department of Computer Science, University of Alabama in Huntsville, Huntsville, AL 35899 United States
AU: Newman, T S
EM: tnewman@cs.uah.edu
AF: Department of Computer Science, University of Alabama in Huntsville, Huntsville, AL 35899 United States
AU: Germany, G
EM: germanyg@email.uah.edu
AF: Center for Space Plasma and Aeronomic Research, University of Alabama in Huntsville, Huntsville, AL 35899 United States
AB: A new shape-based method for segmenting the auroral oval from NASA POLAR Ultraviolet Imager (UVI) data is presented. The POLAR mission has produced millions of UVI images, making automated auroral segmentation a beneficial and critical early processing step in analysis of high-latitude ionosphere-thermopshere-magnetosphere (ITM) coupling using auroral images. Past approaches to automatically or semi-automatically segment the auroral oval from UVI imagery include various types of thresholding, histogram-based K-means, and neural network methods. The existing approaches are generally not robust due to the high noise level, the low level of intensity contrast, and the day glow present in some UVI images. A common shortcoming of existing methods is incomplete detection of the auroral oval for some images. In some cases, existing methods can even fail to detect any part of the oval. The method introduced here is more robust to the challenges of the UVI imagery. Recently, we have demonstrated that the auroral oval's shape in UVI images is well-modelled as an elliptic arc. The segmentation method introduced here exploits this finding; we allow shape knowledge to guide auroral processing. The method involves use of a linear least-squares based shape parameter binning approach that operates on pixels determined from an image-specific thresholding step. The binning approach utilizes a modified randomized Hough Transform scheme that is also fast (faster than conventional binning schemes). The approach treats the inner and outer auroral oval boundaries separately and also incorporates heuristics that allow robust differentiation of appropriate inner and outer boundaries. The new method has been tested on more than 1000 aurora images. Results indicate that the method is highly reliable, even in the presence of high image noise, low contrast, and moderate levels of day glow.
DE: 2407 Auroral ionosphere (2704)
DE: 2704 Auroral phenomena (2407)
SC: SPA-Magnetospheric Physics [SM]
MN: Fall Meeting 2005