ABSTRACT
In our paper (Frackiewicz & Palus 2011) two new methods of initialization were considered. The first method, which is an arbitrary one, is based on uniform partitioning of diagonal of RGB cube (DC) into k segments. Gray levels in the middle of segments are used as initial centres. If an image is clustered into k clusters, k initial cluster centres are located on the gray level axis. The second method, which is an adaptive one, uses a size of pixel cloud of a colour image and the method has beenmarked as SD. First, the mean values and standard deviations (SD) for each RGB component of all image pixels are calculated. Then, around the point of mean colour (a pixel cloud centre) a rectangular cuboid with sides equal 2σR, 2σG and 2σB is constructed. We assume that it lies within the RGB cube. Next, the main diagonal of the cuboid is divided into k equal segments. The centres of these segments are used as initial cluster centres.
