415
Figure 5. Density slices of the ratios and differences. The white
and black areas represent areas of extreme change.
extreme change. These areas, such as the change from grass (low-density)
in the BW scene to water (high-density) in the color scene, mapped out
correctly, while areas of less extreme change in density contained
identification errors. These errors ranged from minor, such as un
changed grass areas being classified as less extreme change, to major,
such as classifying the area of a new building as not having changed.
Although the sun angle and the temporal change problems were still in
fluencing the results, the FA method gave the most promising results.
Five transformations were generated for this method: principal compo
nents transformation, varimax rotation, quartimax rotation, equimax
rotation, and oblimin rotation. Of the five, three rotations contained
useful contrasts between the BW scene and the color scene. Transforma
tion coefficients for the principal components and one selected rotation
are shown in Table 1. Axis 1 of the varimax rotation was the most suc
cessful in mapping change. An example map product is shown in Fig. 6.
The method of density slicing the selected transformed axis into five
levels, used in the third method, was also used in this method. In ad
dition, histograms of the single, transformed contrasting channel data
were used for guidelines in density slicing.
Table 1. Transformation Coefficients
Channels Axes: 1
2
3
4
PRINCIPAL COMPONENTS
TRANSFORMATION
Black-and-white
0.1194
-0.9927
-0.0058
-0.0130
Color 1
0.5752
0.0806
-0.1683
-0.7964
2
0.5713
0.0608
0.7779
0.2544
3
0.5731
0.0653
-0.6053
0.5485
VARIMAX ROTATION
Black-and-white
-0.07048
1.00736
-0.04420
-0.14098
Color 1
0.35905
-0.01958
-1.79843
-8.81860
2
0.30827
-0.02672
4.39597
2.42510
3
0.34721
-0.02395
-2.56193
6.46385