Regardless of the chosen segmentation class, namely region
growing or split and merge, there are two defining
characteristics of the segmentation methods. The first consists
of the similarity distance used to determine the affiliation of a
pixel to a region. Since for color images each pixel is
characterized by three values, the similarity distances used are
the main metrics defined for the general case of vectors, such
as the Euclidean, Manhattan or Minkowski metrics. For the
same color image and with the same segmentation method by
changing the pixel distance will lead in most cases to different
segmentation results. The second characteristic is the color
space used for computing the similarity between pixels.
Changing the color space will lead in most cases to different
segmentation results. For general image segmentation better
results are obtained when using the HSV color space