Graphics Reference
In-Depth Information
2
Image Segmentation
2.1 Introduction
We pay attention to segmentation, as it plays a significant role not only in
image processing but also in pattern recognition. Segmentation of an image is
its subdivision or partition, such that each partition is homogeneous in some
sense. Partitions may be neither geometrically nor physically meaningful, i.e.,
an input image that shows, say, different industrial parts, may not be divided
into regions, each describing one complete physical object (i.e., an industrial
part of its input) or a single geometrically defined object (which means a
completely circular, cylindrical or of any other definition from the input).
Such a segmentation is very dicult and needs semantic knowledge at different
levels of subdivisions, so that division and integration, or the split and merge of
image regions, can successfully exploit this knowledge. Unfortunately, most of
the time we do not have this knowledge. Consequently, segmentation becomes
a dicult task. In the simplest case, one can use the graylevel threshold values
to segment images. Obviously, different segmentations for an input image
are possible, depending on different applications. As an example, segmented
homogeneous regions, along with their contours, may be useful for designing
image compression algorithms, whereas segmentation into known geometric
entities may be useful for industrial inspection and medical diagnosis. A lot
of research work has already been done in the area of segmentation, though
we believe that segmentation still needs attention for semantic partition. An
ideal segmentation or the ultimate objective of segmentation is to separate a
physical object out from a scene.
2.2 Two Different Concepts of Segmentation
Segmentation can be broadly classified into two different groups: contour-
based and region-based segmentation. The idea of segmentation into different
image parts can be viewed as a pixel classification process, where we view the
Search WWH ::
Custom Search