Ww Ww WN 
-~N WW — I UU 
OOC€9 pe MF 0 NO RO o 
EMO -—- I. 
3.2 Vector to Raster Conversion 
Convert the vector strings into a raster representation. 
This data conversion step is relatively simple. Since both the raster 
image and the GIS linework are, presumably, geographically 
referenced, the task at hand is to determine whether a given vector 
falls within a specific pixel. Since we are generally not interested 
in sub-pixel features (ie. features smaller than the spatial resolution 
of the image), such trivial elements should be removed. These 
could be deleted by filtering out or deleting elements less than a 
user-specified threshold. 
Export resulting raster to IAS 
This step is similar to bringing data from an IAS to a GIS. The 
data is translated to an intermediate format, such as DLG, and 
subsequently exported to the IAS. 
The rasterized vectors can then be displayed as an overlay on the 
raw imagery to assess the relative accuracy of the linework. Using 
water bodies, for instance, allows an operator to visually inspect 
whether the linework is geographically accurate with respect to the 
image. In some cases, a ’live-link’ to the GIS database can be 
maintained, but a discussion of this is beyond the scope of this 
paper. 
3.3 Comparison of the two data conversion routes 
In this paper, the accuracy of a feature extraction technique using 
data from a land cover classification was qualitatively compared 
against a rasterization of a GIS vector coverage. Specifically, 
water boundaries were used to reference the two data sets. A 
scheme of scoring both of the procedures based on 4 of the 9 
elements of image interpretation (Bowden and Pruitt (1974)) was 
adopted. The 4 criteria chosen were size, shape, resolution (scale) 
and geometric accuracy of the end products of the processing. If 
the size and shape of each of the elements were similar, a high 
score was given. If the resolution of the elements were closely 
matched, a high score was given. If the elements overlapped well, 
a high score was given for accuracy. The values assigned to each 
criteria were ranked from 1 (poor) to 10 (excellent). The results of 
367 
the qualitative comparison are tabulated in Tables 1 and 2. 
Feature Extraction Technique 
In this case, a control dataset of classified NOAA AVHRR imagery 
that had undergone the feature extraction procedure was used. 
The resulting vector data were imported into the GIS and displayed 
with the manually digitized water body coverage. 
Rasterization of Vector Coverage 
In this case, a control dataset of manually digitized water bodies 
that had been rasterized was used. The rasterized vector data was 
exported to the IAS environment and displayed as an image 
overlay on the unclassified image. The proximity of the raster 
water body theme to known water features was observed and then 
scored. 
Table 1. 
Qualitative Evaluation of the Accuracy 
of a Rasterized Vector Coverage against 
a Georeferenced NOAA image Composite 
  
  
  
  
  
  
  
  
Criteria Performance 
Score (1-10) 
Size 5 
Shape 8 
Resolutio 7 
Accuracy | Svp] 
  
    
  
Table 2. 
Qualitative Evaluation of the Accuracy 
of Vectors Extracted from a Land Cover 
Classification against a Digitized Coverage 
  
  
  
  
  
  
  
of Water Bodies 
Criteria Performance 
Score (1-10) 
Size 5 
Shape 5 
Resolutio 6 
Accuracy LT