International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999
with an incidence angle of 27 degrees). The second corresponds
to the case of a reference map (Michelin Map) and a historic
map (Cassini map of 1756). The local approach guaranties a
residual value of 0 for the tie points used by the process. Table
1 shows the errors (mean, minimum and maximum values)
estimated using control tie points. A considerable improvement
through the use of the local correction is noted due to the fact
that the local deformation is usually smooth.
Polynomial
approach
Polynomial + local
approach
Mountainous area
(34 control tie
points)
19.5,0.5,46.6
9.1,0.7,28.3
Cassini map
(22 control tie
points)
9.1,4.8 , 19.9
6.2,1.1 , 13.1
Table 1. Coregistration errors (mean, minimum, maximum) for
two methods (in pixels).
When a DEM is available (only required for the master image),
we assume that the correction to be applied is a function of the
height (see Eq. 4).
dxi = a Zj + b (4)
dy ; = a’Zj + b’
The coefficients in Eq. 4 are estimated using the set of tie points
(weighted in the same way as previously).
When a DEM is available, this approach can provide more
accurate results. For instance, in the case of the mountainous
area, the error decreases from 9.1 to 1.3 pixels (see Table 2).
Polynomial + local
approach with DEM
Mountainous area
(34 control tie points)
1.3,0.2,2.6
Table 2. Coregistration errors (mean, minimum, maximum)
when using local corractions and a DEM (in pixels).
2.3. Use of Linear Features
Another important characteristic of the GEORIS software is the
capability to use, in addition to the tie points, selected linear
features (road, railway, power line, shoreline, pipe, etc.) to
compute the global and local models. These features are directly
pointed by the operator (depending on the spatial resolution and
the positioning accuracy to be reached, the features should be
carefully pointed at, e.g. either the road strip or one of the edges
of the road or the centerline of the road). When pointing a linear
feature (i.e. road strip), it is not required that the beginning and
the end of the linear feature have the same location in both
images to be registered; only a common segment is needed. The
linear features can be used either to calculate the registration
error (from the results of matching of the linear features after
the global and local models are applied to the image to be
registered) or to compute the models themselves by using
thousand of tie points which are automatically generated along
the selected linear features.
The registration error is estimated using a statistical approach.
For each point of a considered feature, the model is applied and
we search the nearest pixel of the corresponding feature in the
other image. The distance obtained is used as an estimate of the
matching accuracy. To limit matching errors, the matched pixels
which involve the end points of the feature are ignored.
This very simple approach permits to benefit in many cases,
e.g., if only one straight line feature is available, it could be
used by the system. This method is especially interesting to
provide an estimation of the accuracy of the model without
needing a large set of control tie points and with a better spatial
distribution. Table 3 shows results (mean error) obtained for the
previous examples.
Polynomial
approach
Polynomial +
local approach
Polynomial +
local approach
with DEM
Mountainous
area
(ca. 5500 points)
8.3
3.8
1.0
Cassini map
(ca. 2800 points)
3.7
3.4
-
Table 3. Mean coregistration error (in pixels) using feature
points as control tie points.
In this case, linear features are used by the system as control tie
points. Another mode that can be requested from the system is
to use linear features as tie lines in order to improve the model.
This function is implemented by an iterative process which
automatically adds new tie points along a feature. This approach
permits the user to visualise and control the tie points added by
the system.
3. GEORIS TOOL
To meet the user and performance requirements specified, an
exploitation procedure has been implemented to guide the
operator step by step: select the zones of interest; catch the
common features; calculate the registration model parameters;
assess the positioning accuracy; apply the registration model;
archive and export the registered images together with the
registration models and the feature attributes.
On-line easy-to-use tools are available to support the operator to
draw precisely the features with scrollable zoom, improve the
image contrast, adjust the position of ground control points and
assess the relative positioning accuracy of any pixel in the
image to be registered.
As GEORIS is dedicated to prepare a set of images for photo
interpretation tasks, an export facility has been developed to
allow either projection of the raster and vector data in a selected