geometric distortions. In order to remedy this problem, an
iterative geometric correction module has been implemented
in CARIS that allows the operator to interactively improve
the map/image coregistration in a combined display without
having to select GCPs. The operator incrementally
translates, scales and rotates the vector graphics to match
corresponding features in the map and in the raster-image
backdrop. The graphics is redrawn after each adjustment
step. Although the different adjustment parameters are
introduced sequentially, their effects are accumulated to
provide a complete affine (6 parameter) transformation.
Once the desired registration has been obtained, the operator
can edit the map or extract new information from the image
by digitization in the display. (The mouse and standard
CARIS commands are used for these functions.) An inverse
transformation of all new points assures proper registration
with the original digital map base. This operation is now an
integral part of the CARIS Primary Graphics module.
3.2 Off-line Georeferencing
The batch processing version of the map-to-image
registration package in CARIS has three modules.
PERSMAPS transforms points in a map or ground
coordinate system into an image coordinate system.
ORTHMAPS reverses the transformation performed by
PERSMAPS. RECTMAPS can be used for either map to
image or image to map transformation.
PERSMAPS and ORTHMAPS are designed for digitized
aerial photographs. The collinearity equation is the
transformation function. Both programs presuppose that the
camera parameters (coordinates of the exposure station and
the angular orientation of the optical axis of the camera) are
known. This information can be obtained by space
resection, may be available from previous photogrammetric
operations or is provided by an inertial navigation system. A
digital elevation model (DEM) is needed to obtain a rigorous
solution. Failing this, the transformation is based on an
average terrain height.
RECTMAPS is designed for non-photographic images such
as Landsat, SPOT, MEIS, SAR imagery. DEM is not
required. Affine, two-dimensional polynomial up to 5th
order and projective transformations may be selected, as well
as surface fitting with spline functions.
4. CONCLUSIONS
In a fully integrated raster-image/vector-graphics processing
facility, the coregistration of image and map files may also
be accomplished through a map-to-image transformation.
This option is especially attractive when an image is only
used for cartographic editing, map revision or transferring
visually interpreted information into a map file. An inverse
transformation assures that the new information is properly
georeferenced. This approach avoids the potentially
undesirable side effects of resampling on the radiometric
characteristic of the image data and significantly reduces the
computational load. This scheme has been successfully
implemented and tested in the CARIS GIS.
ACKNOWLEDGEMENT
The development work reported in this paper has been
funded under the Canada/New Brunswick Subsidiary
Agreement on Industrial Innovation. The collaboration
provided by Universal Systems Ltd. is also gratefully
acknowledged.
REFERENCES
Derenyi, E. (1991). “Design and Development of a
Heterogeneous GIS.” CISM Journal ACGC, Vol. 45, No.
4.
Derenyi, E. and R.K. Saleh (1989). "Effect of Resampling
on the Geometric and Radiometric Fidelity of Digital
Images." Proceedings of IGARSS'89 Symposium,
Vancouver, B.C., July 10-14, Vol. 2, pp. 620-623.
Table 1: Training statistics before and after resampling.
Field Water
o Res. Mean o Mean o
0 -- 151.4 53 8.1 1.5
9 NN 151.4 5.4 8.1 1.5
: BC 151.2 6.9 7:5 2.1
35 NN 151.5 5.5 8.2 1.5
BC 150.8 8.0 7.4 1.9
& — rotation angle, © = standard deviation
NN - nearest neighbour, BC - bicubic
Table 2: Pixel counts after maximum likelihood classification.
o Res. Field Water
0° we 1532 4110
9 NN 1540 4116
: BC 1876 5020
35 NN 1562 4108
BC 2302 5082
674
