An alternative approach for normalised cuts algorithm to
perform the image segmentation in a more efficient way has
been carried out. With this approach, it enables normalised
cuts algorithms to perform segmentation on image part by
part individually instead of performing segmentation on whole image in one stage. This helps to speed up the
normalised cuts algorithm.
Further improvement can be done by balancing the
trade-off between computation efficiency and effectiveness
such that the segmentation can meet the optimal
performance. Another improvement task is that the number
of image cells to be formed can be adaptively defined
according to the image content. When foreground area takes
larger portion than the background area, the number of
image cells should be cautiously not too many to reduce the
tendency of waste in computation for the local segmentation
in every image cell. Irregular sizes of the image cells can be
formed in an image since the frequency content of the image
is not evenly distributed. For example, region with fine
details suggested to be covered with smaller image cells,
while background that contains fewer details is suggested to
be covered with bigger image cells or less number of image
cells. Nevertheless, the tuning of the image cell size still
needs to meet optimal trade-off without affecting either the
efficiency or accuracy significantly.
A higher level knowledge can be implemented to assist
the segmentation using normalised cuts algorithm. However,
there exists a potential where the use of different primitive
domains to do the segmentation in the absence of high level
knowledge can be done [12].