Reconstruction of the Human Airways - Computational Fluid and Particle Dynamics in the Human Respiratory System - page 57

Region growing is the process where pixels are grouped into larger regions based
on predefined criteria for the growth. To begin, seed pixels are defined which grow
into regions by appending to each seed pixel its neighbouring pixels that satisfy a
predefined criteria (i.e. specific range of grey level). The predefined criteria wield
significant influence on the output. For example specifying pixels to be too similar
gives good homogeneity and coherent regions, but it lends itself to oversegmentation
whereby the region suffers from being smaller than the actual object and probably
won't span over separate objects. More relaxed criteria can produce larger regions
that fill the entire object but may lead to leaks across the boundaries of those objects
and fill multiple objects unnecessarily. Selecting the criteria for DICOM images
typically involves a threshold range on the pixel intensity, or to apply an average
intensity over a neighbourhood around each pixel. For multiple regions, seed points
should be selected within each of the desired regions since the seeded points grow
through neighbouring pixels that are similar, and is expected to end at boundaries.
Region splitting is an alternate method to region growing which begins by divid-
ing up the whole image into disjointed sub-regions and then merges similar regions
together. The splitting criteria continues as long as the properties of a newly split
pair continue to differ from those of the original region by more than a threshold.
Recursively splitting image into smaller regions becomes more efficient than re-
gion growing which involves recursively merging individual pixels to produce larger
coherent regions. The main problem with this region splitting is the difficulty of de-
ciding where to make the partition. One splitting technique is to subdivide the image
into smaller quandrants building a quadtree structure. The process involves looking
at an area of interest to determine if the region satisfies a set criteria. If so, then the
area retains its region, if not, then the area is split into four equal sub-areas and each
subarea is reconsidered with the set criteria. This process continues until no further
splitting occurs. In the worst case the areas become so small that they are actually a
single pixel. One can also allow merging of two adjacent regions if each has similar
characteristics and satisfies the set criteria (Fig. 3.8 ).
The watershed algorithm concept comes from the field of topography, referring to
the division of landscape into water catchment basins or areas. The grey scale image
is seen as a topological surface, where each pixel is a point situated at some altitude as
a function of its grey level. The white colour (grey level 255) is taken as the maximum
altitude and the black colour (grey level 0) the minimum. Deep regions, known as
catchment basins, will exist at local minima, and separated by ridges in the topology
called watershed lines or simply watersheds. Two approaches may be used to extract
the regions of interest. The first is to find basins, then their enclosing watersheds by
taking a set complement. This is analogous to the topolgocial surface being filled with
water. The catchment basins will fill up with water starting at these local minima.
As the flood level rises, boundaries between adjacent segments (catchment regions)
erode and those segments merge. The other approach is to compute a complete
partition of the image into basins, and subsequently find the watersheds by boundary
detection.
This section introduced the reader to some common algorithms that have been
used as a basis for segmentation but is certainly not limited to these algorithms alone.