Page 163 -

CHAPTER
4

Grey-Level Segmentation

4.1 Basics of Grey-Level Segmentation

Grey-level segmentation, or thresholding, is a conversion between a grey-level
image and a bi-level (or monochrome,or black-and-white) image. A bi-level image
should contain all the essential information concerning the number, position,
and shape of objects while containing a lot less information. The essential
reason for classifying pixels by grey level is that pixels with similar levels in a
nearby region usually belong to the same object, and reducing the complexity
of the data simplifies many recognition and classification procedures. Thresh-
olding is almost essential before thinning, vectorization, and morphological
operations.
The most common way to convert between grey-level and bi-level images
is to select a single threshold value. All the grey levels below this value will
be classified as black (0), and those above will be white (1). The segmentation
problem becomes one of selecting the proper value for the threshold T.Since
most grey-level images possess only one byte per pixel, this means that there
are usually only 256 different possible thresholds. Picking the best one could
easily be done by eye but is less trivial to do using an algorithm.
What is being assumed here is that the pixels in an image I belong to one of
two classes based on their grey level. The first class is the collection of black
pixels, which will be given the value one, and for this class:
I(i, j) < T (EQ 4.1)

137