Content-based Retrieval of Auroral Images — Thousands of Irregular Shapes

M.T. Syrjäsuo, E.F. Donovan, and L.L. Cogger (Canada)


Geosciences, pattern recognition, machine vision, content based image retrieval


Contemporary space research uses both satellite and ground-based instruments to measure the auroras. The au roral shapes and their occurrence provide a way to monitor the near-Earth space. However, the ground-based cameras acquire tens of millions of images annually, only a frac tion of which are actually looked at by auroral researchers. Summary data helps, but full utilisation of these powerful data sets demands automated analysis tools. One of the great challenges in developing such tools is that character ising the shapes of auroral forms is surprisingly difficult. The edges and shape boundaries of forms are difficult to deal with, for example, because the aurora is transparent. In this paper, we present a way to extract shapes from au roral all-sky images. We then experiment with commonly used shape descriptors and study their suitability for quant itative searching for auroral shapes.

Important Links:

Go Back