9
overall image entropy in order that it can be converted into vector form which (a) needs a much lower
storage requirement that in raster form, and (b) is readily understood by the end user in conventional
cartographic terms. This generalization process is extremely difficult to carry out. “ (Wilkinson,
1996, p.88)
The goal of the second part of this study was to evaluate whether using high resolution multi spectral
imagery in a GIS processing environment could accomplish the type cartographic generalization that
Wilkinson believes is desired. In order to simulate the type of imagery that will be provided by the next
generation of remote sensing satellites, NASA Stennis developed the CAMS to use with an aircraft (Jensen
et al., 1994). The data have an effective spatial resolution of 2.5 meters and cover nine bands (.42 - 12.5
um). These bands consist of blue to thermal infrared reflectance values. Since the data had been
geographically rectified and projected within an image processing system it could be read directly into the
desktop GIS in the same manner as the SPOT data. Within the GIS any three bands of the image can be
simultaneously displayed using the linear manipulation functions of the legend editor. For example, a
traditional false color image was created by assigning the near IR, red, and green bands respectively into the
red, green and blue image planes of the video display (fig. 7).
2.2.1 Processing CAMS Data as a Grid
The CAMS data were converted directly into nine grid layers with brightness values of 0 to 256.
Using the legend editor, each grid can be displayed separately in monochomatic gray scales. For this
experiment the four most meaningful bands (green, red, near infrared and thermal) were classed into ten equal
area classes (figs. 8-11). In this manner, the response of the different geographic features to different parts
of the electromagnetic spectrum is apparent. For example, it is clear that the streets, running track, and
buildings have high reflectance values in the thermal channel and very low values in the near infrared band.
The major objective of using any remotely sensed data is to convert it into meaningful information.
Traditional air photo interpretation involves the identification of features on the Earth’s surface. This process
typically involves knowing something about what is actually on the ground. The ability to visually identify
features is directly related to the resolution of the image. Since there are 16 CAMS pixels for each pixel of the
10 meter SPOT data, geographic features are much more apparent in the CAMS scene. The initial step in the
process was to identify the brightness values that correspond with known features. According to Jensen (1996),
the red band represents one of the most important bands for vegetation discrimination. Therefore, one may
assume that the dark areas in figure 12 correspond to healthy vegetation that is absorbing red energy. Using
this information it was possible to query selected values on the screen that corresponded to homogeneous dark
areas. This was performed directly through a series of interactive cursor “identify” queries. Once the
maximum threshold value for healthy vegetation was determined the cells in the red grid that were less than
that value were extracted as a new data set (fig. 13).
Given the data format concerns of Wilkinson it was decided to generalize the grid data into a more
conventional vector GIS coverage. The first step in the process was to eliminate the noise in the data.
Essentially the noise results from a “salt and pepper” appearance in the grid. In other words, there are several
isolated cells that would be converted into single pixel polygons. The best way to eliminate this noise is to use
a filter or roving window that examines all the cells in the neighborhood of each cell. According to Tomlin this
is a focal function in the parlance of cartographic modeling (Tomlin, 1990). In this experiment a 3 cell by 3
cell majority neighborhood function was performed. This process filled in the holes in the grid layer and
