1.2.3.4. Object-Based Approaches ( polygon
approach)
By far the greatest advance in classifying digital remotely
sensed data in this century has been the widespread development and adoption of
object-based image analysis (OBIA). Traditionally, all classifications were
performed on a pixel basis. Given that a pixel is an arbitrary delineation of
an area of the ground, any selected pixel may or may not be representative of
the vegetation/land cover of that area. (Gamanya et al.,2008) In the OBIA
approach, unlabeled
7
pixels are grouped into meaningful polygons that are then
classified as polygons rather than individual pixels. This method increases the
number of attributes such as polygon shape, texture, perimeter to area ratio,
and many others that can be used to more accurately classify that grouping of
pixels (Blaschke et al., 2008).
1.2.4. Post-processing
Post-processing can be defined as those techniques applied to
the imagery after it has been through the classification process--in other
words, any techniques applied to the thematic map. It has been said that one
analyst's pre-processing is another analyst's post-processing. It is true that
many techniques that could be applied to the digital imagery as a
pre-processing step may also be applied to the thematic map as a
post-processing step. This statement is especially true of geometric
registration. While currently most geometric correction is performed on the
original imagery, such was not always the case. Historically, to avoid
resampling the imagery and potentially removing important variation
(information), the thematic map was geometrically registered to the ground
instead of the original imagery (Jensen, 2005).
1.2.5. Accuracy Assessment
Accuracy assessment is a vital step in any digital remote
sensing project. The methods summarized here can be found in detail in Green et
al., (2009). Historically, thematic maps generated from analog remotely sensed
data through the use of photo interpretation were not assessed for accuracy.
However, with the advent of digital remote sensing, quantitatively assessing
the accuracy of thematic maps became a standard part of the mapping project.
Once the error matrix is generated, some basic descriptive
statistics including overall, producer's, (Cogalton, 2010) and user's
accuracies can be computed. In addition, there are a number of analysis
techniques that can be performed from the error matrix. Most notable of these
techniques is the Kappa analysis, which allows the analyst to statistically
test if one error matrix is significantly different than another.
1.3. Digital Image Types 1.3.1. Multispectral
Imagery
The dominant digital image type for the last 40 years has been
multispectral imagery, from the launch of the first Landsat in 1972 through the
launch of the latest GeoEye and DigitalGlobe sensors.(Tucker,1985)
Multispectral imagery contains multiple bands (more than 2 and less than 20)
across a range of the electromagnetic spectrum. While there has been a marked
increase in spatial resolution, especially of commercial imagery, during these
40 years it should be noted that there continues to be a great demand for
mid-resolution imagery.
8
| 
