Techniques for searching images from a spectral image database and calculating the distances between spectral images are proposed. The techniques are based on one- and two-dimensional Self-Organizing Map (SOM). For one-dimensional SOM, the Best Matching Unit (BMU) histogram for every spectral image in a database is created, and images of a database are ordered according to the histogram similarity. Two-dimensional SOM is trained by using BMU-histograms as a training data and the distance between spectral images is defined based on their location on the map. The results using real spectral image database are given.
Oili Kohonen, Markku Hauta-Kasari, Kanae Miyazawa, Jussi Parkkinen, Timo Jääskeläinen, "Methods to organize spectral image database" in Proc. IS&T CGIV 2004 Second European Conf. on Colour in Graphics, Imaging, and Vision, 2004, pp 372 - 375, https://doi.org/10.2352/CGIV.2004.2.1.art00074