We know that the pixel value in an image reflects the certain
characteristics of the object in the image. Digital images consisting of pixels
and defined on a bounded 2-D lattice have two important attributes.
First, pixel intensities of a nearly homogenous object/region will follow
a certain statistical distribution. We will call this the conditional (on the
region number) intensity distribution of the pixel intensities. The pixel
intensities may be scalar or a vector quantity depending on the image
type (see Chapter 2). Second, the pixels close together or lying in a
neighborhood will tend to have similar intensity values. The latter constraint
is also known as contextual information. Intensity-based segmentation
methods that we discussed in Chapter 6 disregard the latter attribute
of images. In this section, we will discuss how contextual information
can be modeled by means of MRF modeling. This model representing
the local characteristics of the underlying image, when combined with
the conditional distribution of the pixel intensities under a Bayesian