Aurora image segmentation by combining patch and texture thresholding
Computer Vision and Image Understanding
| Hi-index | 0.00 |
The auroral emissions observed in the high-latitude regions encircling the magnetic poles are a key element in studying plasma-physical processes in the near-Earth space, the magnetosphere. The Finnish Meteorological Institute operates five digital all-sky cameras, which routinely monitor the auroral emissions in Northern Finland, Sweden, and Svalbard; each camera records an image of the full sky at 20-s intervals.In this paper, we develop a method that allows us to examine such a large data set by classifying the images through determining the shape skeletons of the auroral forms in each auroral image. Shape skeletons are a commonly used representation of object shapes in machine vision applications. Once determined, shape skeletons have the advantage that they can also be used to represent noisy or unevenly distributed data. Here we apply a skeletonizing algorithm to determine the skeletons of auroras in a noisy environment. The algorithm is based on batch mode self-organizing map. The results can be further improved by implementing understanding of the auroral physics to the algorithm.