Building image icons for fast browsing and classification of solar radio spectrograms

We address the problem of how to retrieve information from data recorded by scientific measuring systems. Possibilities are required to help interactive searches in large image databases. In particular, tools to (1) browse quickly among large images (MPixels) and (2) search for similar images are to be developed. We propose to achieve this demands by building an additional, smaller representation of the image. We describe how a vectorial representation can be constructed by partitioning the three-dimensional attribute space of the image. The resulting image icon not only contains the most important features of the primary image, but can also be used to search for similar images in the database. The method is mainly applied to non-spatial images, e.g. measurements of some multidimensional, physical parameters. It is illustrated on images called solar radio spectrograms which measure the intensity of the solar radio flux in a given time vs. frequency