An image segmentation algorithm based on irregular partitions of the image
is presented which avoids some of the classical drawbacks of common hierarc
hical segmentation techniques. A reduced number of hierarchy levels are obt
ained by applying a new clustering technique at every level using the neare
st neighbour criterion. The resulting clusters are connected and no relinki
ng operations are needed. The algorithm has been tested on grey scale image
s, giving excellent segmentation results and involving a low computational
load.