two files; 
naximum 
'estricting 
iber(s) or 
| the same 
| Specify a 
S Straight 
Joints) or 
smoothly. 
1tine were 
ly 60% of 
ince must 
there will 
ng. 
ours 
0 data set 
) 000 data 
the metric 
I's policy 
r all new 
val would 
ntour line 
data set. 
or the re- 
had been 
rs and to 
ly derived 
at must be 
specified. 
procedure 
e derived 
odes were 
its of each 
d at the 
- content, 
sentation. 
ach other 
ts of the 
map sheet 
favorable 
ralization 
s. 
zery 
that was 
bands and 
1:250 000 
olution of 
ided and 
ediate file 
  
  
  
  
format) by Canada Centre for Remote Sensing 
on the PCI System and input to the CARIS 
System. The CARIS System then reformatted 
the PIP file to the CARIS IPV format which 
was required for processing and display. 
3.2.1 Measurement and Evaluation of 
Reference Data 
In preparation for the measurement of the 
reference points, spectral bands 7, 2 and 1 of 
the Landsat TM image were selected and the 
image was enhanced by performing contrast 
stretching through a histogram equalization on 
the red, green and blue bands. This enhanced 
the detail in the image and improved the 
identification of the selected control points. To 
collect data from a Landsat image, the image 
must first be registered to projection reference 
system to ensure that collected data is in its 
proper planimetric position. Due to the small 
mapping scale, low relief and immediate 
unavailability of DEM, correction for relief 
displacement was not performed for the TM 
image. The registration of the image to the 
vector data was performed by a polynomial type 
transformation (Welch et al., 1985; 
Colvocoresses, 1986). 
Thirty-five (35) reference points were measured 
interactively in both the vector file and Landsat 
raster file. In some cases, digitizing these 
reference points proved to be difficult, 
especially where the point was located in an area 
of similar spectral reflectance such as a rural 
road intersection surrounded by cultivated 
fields. An interactive evaluation of the control 
and reference points was performed. This 
functionality allowed two data sets to be 
graphically displayed and their respective 
control files to be evaluated during the 
adjustment. When evaluating the control 
points, the operator may remove points with 
large residuals, add new points, re-run the same 
adjustment or perform the adjustment using 
another transformation. The geometric 
correction of the Landsat TM image was 
performed using a third degree polynomial 
transformation. Twenty-one (21) control points 
were selected for the rectification of Landsat 
image and the registration to the vector data. 
The standard deviation of the planimetric 
residuals after the adjustment was t14.7m 
(£10.9m in x and +17.7m in y). Nearest 
neighbour resampling was applied to the image 
and the output pixel size was kept at 30m. To 
obtain an indication of the quality of the 
adjustment, the coordinates of fourteen (14) 
image check points were measured compared to 
their "true" ground values. The standard 
deviation of the differences between "true" and 
measured coordinates were calculated and for 
ten of the check points were +12.6m in x and 
+16.1m in y. The coordinate differences in the 
other four check points varied between sub- 
pixel values to two-pixel values. These results 
satisfy the NATO A rating planimetric accuracy 
(CMAS-125m) for the 1:250 000 maps, 
equivalent to standard deviation of differences 
of +41.2m. 
3.2.2. Change Detection and Vector Data 
Collection 
The task of change detection was performed 
visually by displaying the integrated vector and 
raster files to determine areas within the vector 
file that have experienced physical change. 
Only minor changes, such as the enlargement of 
a couple of gravel pits and the addition of a road 
were detected. To provide an opportunity to try 
the revision functionality, features from the 
vector file were masked to simulate change. 
For the collection of the 'updates', a unique 
layer number was set to differentiate the data 
collected on the Landsat TM image from the 
original vector data. This was necessary for 
later separation of the Landsat data for 
comparison to the original data. Data collection 
was performed interactively using screen 
digitization (mouse & cursor). The features 
collected were: roads, built up areas, water 
courses, waterbodies, gravel pits, forests and 
power lines. The identification of features was 
considered biased, as their identity was already 
known from the masked data. It was found that 
in areas of high contrast the feature delineation 
was highly accurate. Examples of features that 
were relatively easy to interpret were power 
transmission lines cut through forested areas, 
water body boundaries (but not narrow 
streams), most forest areas, gravel pits and 
multi-lane roads. Classification of roads and 
the differentiation between cut lines, roads and 
power lines would not be possible directly from 
a Landsat TM image and would require field 
verification or the use of ancillary data. Cases 
where it was difficult to determine the feature 
delineation were between rural roads and 
cultivated fields, built up area extent, residential 
roads through built up areas, and between some 
forests and fields. Since digital aerial 
photography was available, it was referenced to 
291