EM 1110-2-2907
1 October 2003
transforms the data using a form of factor analysis (eigen value and eigen vector matrix).
For a complete discussion of the technique see Jensen (1996).
Figure 5-15. PC-1 contains most of the variance in the data. Each succes-
sive PC-transformation isolates less and less variation in the data. Taken
d. Image Classification. Raw digital data can be sorted and categorized into thematic
maps. Thematic maps allow the analyst to simplify the image view by assigning pixels
into classes with similar spectral values (Figure 5-16). The process of categorizing pix-
els into broader groups is known as image classification. The advantage of classification
is it allows for cost-effective mapping of the spatial distribution of similar objects (i.e.,
tree types in forest scenes); a subsequent statistical analysis can then follow. Thematic
maps are developed by two types of classifications, supervised and unsupervised. Both
types of classification rely on two primary methods, training and classifying. Training is
the designation of representative pixels that define the spectral signature of the object
class. Training site or training class is the term given to a group of training pixels. Clas-
sifying procedures use the training class to classify the remaining pixels in the image.
5-29
Previous Page
