Image Segmentation - Bezier and Splines in Image Processing and Machine Vision - page 34

2.2.2 Region-based Segmentation

Region-based segmentation mainly depends on either thresholding or region
growing, merge, and splitting. Selection of thresholds has an important role
in threshold-based segmentation, e.g., single level thresholding produces a
partition in the way that
f ( x, y )= f ( x, y ) when f ( x, y )

≥ T,

(2.3)

0 otherwise,

where f ( x, y ) is the gray value in the image and T is a threshold. There
are many ways by which one can calculate T. The simplest way is to use
the histogram of the image. Equation (2.3) provides binary segmentation or
object/background segmentation when f ( x, y ) is taken as 1 for f ( x, y )

≥ T

and zero otherwise.
For multilevel thresholding, we choose
f ( x, y )= f ( x, y ) when T i ≤

f ( x, y ) ≤ T i +1 i =1 , 2 , ··· k,

(2.4)

0 otherwise .

By multilevel thresholding we can separate out different segments of an image
corresponding to different ranges of gray values. This corresponds to different
objects or different portions of an object in an image.
Recursive thresholding can also be used for good segmentation. For this,
segment an image corresponding to a threshold and if that segmentation
does not fulfill certain objectives, then re-segment the segmented image
through an iterative computation of a new threshold. So, segmentation and
re-segmentation go on continuously until the criterion is satisfied for a definite
task.

Region Growing, Merge, and Split

Region growing normally starts from a small region, and merges small nearby
regions to grow in size. If the merge is successful, neighborhood regions are
further merged depending on a condition for successful merge. The process
can keep on running if the merge passes the test, otherwise, the merge is
declared unsuccessful, and split of the previous merge is carried out.

2.3 Segmentation for Compression

We now discuss how we can obtain a good segmentation of an image for image
data compression. Choose the region-based segmentation of an image rather
than the contour-based segmentation. Region-based segmentation is more use-
ful and effective in image data compression because region contours are not
disconnected like edges. Keep in mind that a contour-based segmentation may