Normally, the classification generates an image with a lot of noise, caused by isolated pixels or by the
few pixels attributed to several classes, that occur near homogeneously classified areas (Crósta, 1993, p 129). With this, it is necessary to homogenize the classification result based on dominant classes.
The Post Classification should eliminate isolated points classified differently from their neighborhood. This results in a classified image with a less noisy appearance.
In a window of 3 x 3 pixels, the center point is evaluated as to the frequency of the classes in its neighborhood.
According to user-defined weight and threshold values, this center point will or won't have its class replaced by the highest-frequency class in the neighborhood.
The weight ranges from 1 to 7 and defines the number of times that the center point frequency will be considered.
The threshold also varies from 1 to 7 and is the frequency value above which the center point is modified.
The lower the weight and the lower the threshold, the greater the number of replacements that will be performed.
It is
accessible through:
PROCESSING → RASTER PROCESSING
→ POST CLASSIFICATION...
Selection
of the layer to apply the post classification
On the Layer field select the raster layer to apply the post classification.
Post classification parameters
Set the value of weight and threshold.
Check in the Preview box to activate visualization of operation and draw the area to be used by the post classification operation.
Output
information
Repository -
Select one type of repository by
clicking on the
button to save the output layer as a file or on the
button to save it in the database.
Layer Name - inform
the raster name.
Press the OK button
to save the resulting.
Hint: The resulting image will be added as a new layer at the
TerraView project.