Techniques for searching images from a spectral image database and calculating the distances between spectral images using different distance measures are proposed and the importance of the normalization that is based on the human visual sensitivity function is examined. The searching techniques are based on the use of one-and two-dimensional Self-Organizing Map (SOM). In the case of one-dimensional SOM, the Best Matching Unit (BMU) histogram is created for every spectral image of a database, and images are ordered according to the histogram dissimilarity. Two-dimensional SOM is trained by using spectral data and BMU-histograms as a training data and the distance between spectral images is defined based on the histogram dissimilarity and image locations on the map, respectively. The proposed techniques are useful in image search and the order of the database is different for spectral images and for spectral images weighted by human visual sensitivity function. The order of the database is also highly dependent on the used distance measure. The results using a real spectral image database are given.
Oili Kohonen, Timo Jääskeläinen, Markku Hauta-Kasari, Jussi Parkkinen, Kanae Miyazawa, "Organizing Spectral Image Database Using Self-Organizing Maps" in Journal of Imaging Science and Technology, 2005, pp 431 - 441, https://doi.org/10.2352/J.ImagingSci.Technol.2005.49.4.art00014