International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
e the minimum correlation between homologous points
(default value is 0.70).
The “outlier rejection algorithm” frame is used to specify the
method to be applied to remove blunders from the set of
computed CPs; an affine or a 2" order polynomial transform
may be selected.
Finally, the “georeferencing parameters” frame is used to set a
minimum number of valid CPs (default value is 8) and to select
the final geometric transformation model (transformation type).
At the moment, only affine and 2" order polynomial
transformations are available in the GEOREF environment.
Anyway, more sophisticated geometric models (higher order
polynomials, RFM, splines, efc.) can be computed with third
parties remote sensing software using GEOREF outputs in
ASCII format.
A set of configuration parameters is expected to work well on a
set of images which are similar for type, content and size, then
it can be saved into a template project to be used at different
times.
Project configuration parameters Seas EM E Bh x]
*Fortinet opiétalor- ^oi REA Matching pacsmeters— Wu DEN,
N. level mas [10 N races ls |
Half size Forster window. [2 Hat window soe amtinglevet [3 1
Distance min. Forsinen EL . Half size search vandow. E
Exculasty Index Mm: fo 3 Mav size search window: [s
Min. points for square [100x100pe:} IE Miriroum corelabon: |0 7
Points density: [0.00500 Fiejection test: B
Maniterations [0e —
—Georef start affinis Iransform coetficienis —————- —-——34 -
beef Sp
300: [92.567550 201; [05351116 02: 0.073823 Si Slay
F waipulondgalm — — — —]1
J 587869 + A11: [0.0673 [0.983945% 1
ai0 | 345.87869 A11:100679494 a12:|0.93384533 | € LSMAffre © (SM Polynomial |
1
"Logfile 1 n7 7p; Beoteferencing parameleis m m mm eee
c Without log fie (Fast) Transformation Wel en
£^ "With log He [Slow] Minimum yald points: [8 00
pz
cr |
Figure 2. GEOREF configuration parameters window.
2.2.2 Interest point extraction and matching: After setup
of parameters, GEOREF proposes an operational workflow to
extract CPs:
a) generation of image pyramids: this task is
implemented to cope with all cases where bad parameters
for the starting affine transformation have been computed.
However, the current version of the software is provided
by a graphical tool which allows to manually measure with
ease a minimum set of 3 CPs to compute this
transformation. Findings of recent tests have shown that
the use of image pyramids could be avoided in the most
cases, working directly on the original images;
b) extraction of IPs: the user is requested to enter the
maximum number of IPs to accept in order to reduce the
computing time; in this case, all extracted IPs are ranked
according to their interest value, and those featuring higher
positions are selected (according to the strategy proposed
in Forlani et al., 1996). Ohterwise, all IPs can be held;
c) measurement of the initial affine transformation,
d) finding corresponding points: the initial affine
parameters are used to transfer each IP from master to
slave image; around the position found on the s/ave, all IPs
falling into a square searching window (see 2.2.1) are
matched with the point in the master. That point featuring
the higher correlation after L.S. matching is selected as
homologous; however, correlation must be higher than the
minimum acceptance threshold. In case not enough points
have been found, the process is repeated by enlarging the
size of the search window.
e) outlier rejection: to cope effectively with blunders in
the set of CPs, a robust technique based on Least Median
Squares (Rousseeuw & Leroy, 1987) followed by classical
L.S. estimation and data snooping is applied.
2.3 Visualization of results and data export
[n case the georeference has been successfully completed, the
user can check results by means of either analytical and
graphical tools. The former concern the sigma nought of the
estimated transformation (affine or polynomial); the
implementation of plotting residuals onto the image is ongoing.
Tha latter is given by overlapping the resampled slave image
onto the master, so that the accuracy of the co-registration can
be verified. An example of this visualization is presented in
Figure 8.
Different kinds of information can be exported from GEOREF.
First of all, a file containing coordinate of CPs in both images is
available. This file could be then imported into other Remote
Sensing data processing softwares, in order to recompute the
registration with more accurate algorithm, based on the same
CP set found by GEOREF.
In case the master image is provided by an ESRI “world” file,
also that of slave is written; this fact enables to directly import
the slave image into a commercial GIS software, so that a
spatial database could be easily upgraded.
Furthermore, all intermediate files and images are available as
by-products and can be used independently from GEOREF.
3. DATASET
GEOREF has been tested with data collected by different high
resolution image sensors (Eros-Al, QuickBird and SPOT-5
HRG). The images used in this study were taken over the
Caselle airport, near the city of Torino, Italy (Figure 3) and the
full dataset used is composed of:
e one scene acquired by the QuickBird satellite;
a one scene acquired by the SPOT-5 satellite;
e one stereo scene acquired by the Eros-A1.
The main purpose of the tests has been to show the potential use
of GEOREF for HRSI automatic registration and for multi-
resolution and multi-sensor automatic data fusion. Therefore,
tests have involved:
e 2.5-meter SPOT-5 HRG and 0.62-meter QuickBird PAN
image fusion (multi-resolution and multi-sensor data
fusion);
e 2.6-meter Eros-Al stereo coverage (HRSI automatic
registration);
e 2.6-meter Eros-Al and 2.5-meter SPOT-5 HRG image
fusion (multi-sensor data fusion).
384
Internat
Mie n
Figur
4.1 Qi
QuickE
m GSL
25° off
The BI
high-re
SPOT-
sensors
range
acquire
the da
geomet
A firs
registre
SPOT
Caselle
m row
image |
Using
automa
transfo
fusion
6.
42 E
Eros-A
images
1.5 m
satellit
CCD :
satellit
collect
differe
À secc
deforn
taken
GEOR
GEOR
overla;
results
origin:
betwec