In this paper we propose a both spatial and colorimetric distance D for a two dimensional color pallet built from the baker's transformation. The baker's transformation provides a quantization of the image into a space where colors that are nearby in the original space are also nearby in the output space, thereby providing dimensionality reduction and invariance to minor changes in the image. Whereas, the distance D provides for partial invariance to translation, sight point small changes and scale factor. Our feature is used in an image retrieval process that has to help a missing robot in an indoor environment. We built a structured image database corresponding to the flat where the robot works. When the robot is lost, it takes an image of its environment that we call request image and the system looks for the closest image in the database witch indicates its position (room and orientation). A hierarchical approach is then conceived by describing in the off line phase each room with a color feature. In a first search phase, we eliminate rooms whose colors are far from those of the request image and then we achieve the search by our distance D. Results obtained with this approach are better than those of the classical color histograms.
A. Chaari, S. Lelandais, C. Montagne, M. Ben Ahmed, "Color image retrieval techniques for a global localization of an indoor mobile robot" in Proc. IS&T CGIV 2006 3rd European Conf. on Colour in Graphics, Imaging, and Vision, 2006, pp 189 - 194, https://doi.org/10.2352/CGIV.2006.3.1.art00038