Colour Space Selection for Unsupervised Colour Image Segementation by Analysis of Connectedness Properties

In this article, we propose a new colour image segmentation algorithm by unsupervised classification of pixels. This procedure iteratively constructs the classes by histogram multithresholding. For this purpose, the procedure selects different colour spaces in which the pixels belonging to the modes of the 1D-histograms present remarkable connectedness properties, so that each mode corresponds effectively to a region in the image.