Image Processing Reference
In-Depth Information
•
ROI collection
: You will now generate a number of training samples to supervise
the classification. These training areas are called regions of interest (ROIs).
From
Basic Tools
(on the main horizontal tool bar at the top of your screen
choose) choose
Region of Interest
and then
ROI Tool
. The new window will
record all of your training areas as you collect them from the image (because of
the size of the subset you may want to use the zoom window to navigate). Click
on the
*
next to
Region #1
to highlight it, then begin drawing a polygon by mov-
ing the cursor around the area of interest, clicking the left mouse once to draw
an arc and clicking the right mouse once to finish the polygon, clicking right
again will record the ROI permanently. Collect training areas for four types of
residential configurations in Figs.
8.5
and
8.6
as well as for the commercial area
labeled as E and the school in the north-west of the subset. Every time a new
training area is collected you will need to click on
New Region
. An erroneous
ROI can be deleted.
•
Classification
: You are now ready to classify. Again, from the main horizontal
tool bar at the top of your screen choose
Classification
and
Supervised
and then
Maximum Likelihood
. Select the
belfast_subset
image
.
Click
OK
. Another win-
dow will list the
Maximum Likelihood Parameters
.
Select All Items
. Click on
Multiple Values
and then
Assign Multiple Values
the following prior probabili-
ties to each
Region
by highlighting it and typing in the
Edit Selected Item
:
Assign probability threshold
Region
Probability
A: Very high density, strong linear residential
0.25
B: Low density, weak linear residential
0.21
C: Medium density, strong linear residential
0.31
D: Very low density, weak linear residential
0.14
E: Low density, weak linear commercial
0.07
X: Single commercial
0.02
Click
OK
when done. Do not click
OK
on the main
Maximum Likelihood
Parameters
window, yet.
•
Experiment:
In order to appreciate the power of variable prior probabilities you
can
Preview
the classification every time you change the probabilities. Experiment
with combinations of prior probabilities to see the effects. When ready, you may
generate a full classification by choosing an
Output Result File
name. In the
Available Bands List
select your classified file and
Load Band
in
New Display
.
From
Tools
,
Color Mapping
and
Class Color Mapping
you can change the col-
ors of the classified image. Hopefully, you will have finished within the 7 min!
Acknowledgements
ADDRESS-POINT
TM
and COMPAS
TM
are Crown Copyright (www.ordsvy.
gov.uk and www.osni.gov.uk respectively). Cities Revealed is the copyright of the GeoInformation
Group (www.crworld.co.uk). IKONOS image was provided by Space Imaging. The author would
like to thank Paul McKenzie for the production of Fig.
8.4
.