Remote Sensing Reviews: To Cite This Article: A. Bannari, D. Morin, G. B. Bénié & F. J. Bonn (1995) A Theoretical
Remote Sensing Reviews: To Cite This Article: A. Bannari, D. Morin, G. B. Bénié & F. J. Bonn (1995) A Theoretical
To cite this article: A. Bannari , D. Morin , G. B. Bénié & F. J. Bonn (1995) A theoretical
review of different mathematical models of geometric corrections applied to remote sensing
images, Remote Sensing Reviews, 13:1-2, 27-47, DOI: 10.1080/02757259509532295
Taylor & Francis makes every effort to ensure the accuracy of all the information
(the “Content”) contained in the publications on our platform. However, Taylor
& Francis, our agents, and our licensors make no representations or warranties
whatsoever as to the accuracy, completeness, or suitability for any purpose
of the Content. Any opinions and views expressed in this publication are the
opinions and views of the authors, and are not the views of or endorsed by Taylor
& Francis. The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and Francis
shall not be liable for any losses, actions, claims, proceedings, demands, costs,
expenses, damages, and other liabilities whatsoever or howsoever caused arising
directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes.
Any substantial or systematic reproduction, redistribution, reselling, loan, sub-
licensing, systematic supply, or distribution in any form to anyone is expressly
forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
Remote Sensing Reviews, 1995, Vol. 13, pp. 27-47
Photocopying permitted by license only
© 1995 Harwood Academic Publishers
Printed in Malaysia
ABSTRACT
1. INTRODUCTION
27
28 A. BANNARI ET AL.
of these two types of data requires a richer exploitation of satellite imagery and
sounder land management policies. The combination is however possible only if
the geometry of all of the documents is identical and if the pixels of each grid
can be superimposed perfectly. Although most of the documents used in geo-
graphical information systems do not present any difficulty since they are derived
from a cartographic base, it is not the case for remote sensing images. Their
geometry derives from the acquisition mechanisms used and often appears far
removed from established cartographic norms (Caloz et al., 1991). Only after
applying adequate geometric correction do many applications and uses of high
spatial resolution digital images become possible, such as:
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
from the same principles pertaining to the condition of collinearity but is related
to celestial mechanics. T h e coefficients of this system of equations are a combi-
nation of different variables relating to the osculating parameters of the orbit, to
the parameters of the viewing geometry and the geographical position of the im-
age centre which intervene during the transformation procedure. The third model
based on a first, second or higher order polynomial is less precise than the two
other methods.
During scene acquisition, the earth turning at an angular speed (ye) from west to
east, the first line (i) will always be more forward in relation to the next (/ + 1)
by a value equal to Ax towards the west. The resulting image has thus the form
of a parallelogram rather than a rectangle (Figure 1). The displacement between
the image lines is variable according to the latitude. It is maximum at the equator
where its value reaches 465 m per second for the French SPOT satellite while the
LANDSAT satellite, for covering a ground track of 185 km takes approximately
26 seconds during which the earth turns 7 minutes in longitude towards the east.
The result is a progressive displacement of each of the scan lines; the last line
of an image at the equator can be displaced by approximately 8 km in relation
to the first. This distortion is so important that the organizations responsible for
marketing satellite images partially correct images from this type of error before
providing raw images to users by displacing systematically, in the reverse direc-
tion, the image lines following specific satellite movement conditions (Rochon,
1982). The resulting error can thus be modelled as follows (Richards, 1986):
Axe = vets (1)
with
ve=WeRezos(X) (2)
where
BandN°l ^
L Image\ Uncorrected image
I1 corrected^
-Ik
Ax L
for
earth
. curvature
' I
^_ BandN° n+1
Ax
FIGURE 1 The effect of earth rotatioa
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
The preceding error takes into account only earth rotation at the time of scene
acquisition. In reality, we must also take into consideration earth curvature, be-
cause this phenomenon becomes sensitive only for large viewing angles. Contrary
to aircraft scanning systems, spaceborne imaging systems introduce a significant
additional distortion in the image caused by the earth curvature. For NOAA
iWHRR with a swath width of 2700 km and an altitude of 833 km, for example,
this anomaly is most important at the edges of the image. The increase in pixel
size can be computed by referring to the geometry of Figure 2. The effective
pixel dimension in 'the inclined earth's surface is evaluated using the following
formula (Richards, 1986):
Pc = p{h + Re[l - cos(0)]}sec(0)sec(0 + 6) (5)
where
Pc is the pixel at the extremity of the image deformed due to earth curvature,
P is the sensor view angle (rad),
6 is the scan angle,
h is the satellite altitude,
(p is the angle subtended at the centre of the earth.
Each point of the earth surface is characterized by an altitude above the mean
sea level. The viewing geometry of the satellite towards a point on earth present-
ing varying altitude values entails a displacement of the position of this point in
relation to its position in the cartographic projection system chosen during the
geometric correction. This displacement is called "parallax" in photogrammetry;
the more important the relief, the higher the parallax. Figure 3 clearly outlines
the effect of terrain relief when passing from the original image to the ortho-
GEOMETRIC MODELS APPLIED TO RS IMAGES 31
Satellite
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
Earth surface
Earth centre
FIGURE 2 The effect of each curvature.
image. The intersection of the field of view with the ground produces pixels of
variable size following the slope and aspect. According to Caloz et al. (1991),
the effect of relief causes a variation in the distance separating two consecutive
pixels along the scan line. For correcting terrain relief effect, it is necessary to
have an altitude value for any point on the image, namely, a digital elevation
model (DEM). The sampling step of this model must be in the size range of the
image to provide an ortho-image with good precision (Novak, 1992).
Satellite altitude over the earth surface is determined by three distinct phenom-
ena:
— the flattening of the globe (the polar radius is less than the equatorial radius),
— orbit eccentricity,
— the relief of the earth surface, i.e. elevation differences over the areas over-
flown in relation to the geoid.
32 A. BANNARI ET AL.
Satellite
.Topographic surface
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
eference ellipsoid
Ax = C-7— (6)
where
Ah is the satellite altitude variation,
C is a constant,
hn is the nominal altitude of the satellite in relation to the earth surface.
Satellite attitude variation during image acquisition causes rotations and transla-
tions of the image lines around the coordinate system related to the satellite. The
GEOMETRIC MODELS APPLIED TO RS IMAGES 33
I
t •Ay
+Ay
a +Ax
FIGURE 4 The effect of satellite altitude variation.
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
dx
Yaw Pitch Roll
(a) (b) (c)
FIGURE 5 The effect of satellite attitude variation.
effect of attitude variation can be summarized into three errors: roll, pitch and
yaw.
During image scanning, variations in attitude manifest themselves either by
a modification of the position of the image centre in relation to the planned
position (variation in roll and pitch), or by a modification of the deformation
aspect due to the earth rotation phenomenon, i.e. by a modification of the angle
between the scan lines and the satellite ground track (variation in yaw), (CNES,
1986).
If the yaw angle remains constant, the effect produces a small change in the
angle between the direction of the scan lines and the satellite ground track. The
rotation angle of the lines is equal to the yaw angle considered constant (dx)
(Figure 5a). The effect of constant yaw hence occurs in relatively the same man-
ner as the earth rotation effect. For vertical scene acquisition in the vicinity of
the equator, for the SPOT satellite, this effect is approximately 25 times weaker
than the one measured for earth rotation. However, a constant pitch angle (dcj))
produces a simple along-track translation in the planned scene centre at a de-
termined date (Figure 5b). In the case of SPOT, this displacement remains less
than 700 m (CNES, 1986). For a constant roll angle (du), this effect translates
itself into a dephasing between images lines parallel to each other. The effect
34 A. BANNARI ET AL.
— roll:
dy = h(l + £yu (9)
where dx, d<p and du are given in radians.
dy = C— (10)
where
Av is the satellite variation,
vn is the satellite nominal speed,
C is a constant.
GEOMETRIC MODELS APPLIED TO RS IMAGES 35
When departing from the vertical look for oblique scene acquisition, the in line
sampling step increases progressively because of the increase in the look distance
and the inclination of the earth surface on the optical axis of the instrument. The
displaying of the image in a uniform grid shows a "lack" of pixels which are more
and more compressed from the centre towards the image extremities.
For the sensors used in satellite or airborne remote sensing, the field of view
of the sensor "{}", the scanning angle "6" (case of the Landsat-TM and MSS sen-
sors) and the inclination angle for oblique viewing (case of SPOT) are constants.
Figure 7 shows that the ground pixel size is larger at the extremities of the scan
line than at nadir. For systems with larger fields of view, such as NOAA AVHRR,
the effect can be quite severe. Let P be the pixel size at nadir and Pg the pixel
size in the scan direction under an angle 6. The following formula shows the
relationship between the two dimensions P and Pe (Richards, 1986):
Pe = P[sec(0)]2 (11)
where
P is the pixel dimension at nadir position,
Pe is the pixel dimension at a scan angle 6.
The sensors which record images line by line using linear array detectors have
a constant ground track if the satellite speed remains constant. As to the sys-
tems which use an oscillating mirror, such as the Landsat Multi-Spectral Scan-
ner, these incur some nonlinearity along the scan line near the boundary of the
ground track. This error is due to a reduction in the mirror rotation speed at the
time of scan direction change, along a line. This effect is represented in Figure
8 (Colwell, 1983). According to Anuta (1973), as cited by Rochon (1982), this
error can cause a displacement of 395 m in pixel position in comparison to
the position this same pixel would occupy if a perfectly linear scanner had been
used.
36 A. BANNARI ET AL.
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
P[sec{ef
Inclined position
FIGURE 7 Effect of panoramic distortion (according to Richards, 1986).
S
o
Track deviation
The geometric correction of remote sensing images consists, above all, in es-
tablishing the relationship between the image coordinate and ground coordinate
systems. For applying these transformations, various methods can be used. Three
methods were investigated in this paper:
— the method based on the equations of collinearity (Slama, 1980; Colwell,
1983; Shu, 1987),
— the method based on the equations of collinearity related to celestial mechan-
ics (Guichard, 1983 and 1985; Toutin, 1985 and 1986; Toutin et al., 1989 and
1991),
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
Ground reference
FIGURE 9 Cojlinearity geometry: case of the acquisition of a vertical aerial scene.
where
When the two vectors r and R (Figure 9) are collinear, the following relation-
ship occurs:
Xp — Xo
= KM Yp-Yo (12)
where
K is the scale factor defined by the quotient of the norms of the two vectors r
and R.
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
where
mu = cos(0)cos(x)
= cos(w)sin(x) -
m2l = -cos(</>)sin(x)
m22 = cos(w) cos(x) - sin(w) sin(^) sin(x) (14)
= sin(w)cos(%) + cos(w)sin((/>)sin(x)
= -sin(w)cos(^)
= cos(w)cos((/>)
By calculating the matrix product, we obtain the three following relationships:
x = K[mn(Xp - X0) + m12(Yp - Yo) + mn(Zp - Zo)]
y = K[m21(Xp - Xo) + m22{Yp - Yo) + m2i{Zp - Zo)} (15)
- Xo) + m32(Yp - Yo) + mi2(Zp - Zo)].
If the first and second equations of relationship (15) are divided by the third
equation of (15), we obtain the fundamental collinearity equations:
= — and A= (17)
.Zo_
xp Xo Up' 'AX1'
YP = Yo XABC V
P + AY' (19)
1
zP. .ZQ. .WP. AZ
where
A is the satellite attitude matrix in relation to the ground,
B is the rotation matrix of the instrument reference in relation to
the satellite reference,
C is the rotation matrix of the sensor system in relation to the
instrument system,
[up,vp,Wp] are coordinates of the image point in the reference (w, v,w),
(AX', AY', AZ') are coordinates of the origin of the instrument reference in the
satellite system (translation).
Following Figure 9, we can write:
Up X
V
P = y (20)
-WP.
It must be noted that in the case of photogrammetry, the two matrices B and C
are equal to the unit matrix /:
GEOMETRIC MODELS APPLIED TO RS IMAGES 41
By defining:
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
XD X0
Yo + XA (23)
Since the image is recorded line by line, we can conclude in these conditions that
equation (23) is valid only for one scan line, which explains the null value of x.
The coordinates XQ,YO,ZQ and the attitude angles w, (f> and x contained in matrix
A correspond only to line (j) of the image acquired at the time tj (Shu, 1987).
The different image lines are recorded for different satellite positions, thus with
different attitude angle values. Moreover, line (j) has for coordinates Xj on the
image:
xj = (tj-to)us (24)
to being the time of acquisition corresponding to the starting line and us the
speed of movement of the satellite. When the viewing geometry for this type of
sensor is vertical, relationship (23) remains valid. But if the viewing geometry is
inclined in relation to the vertical of an angle "a" in the cross-track direction
(case of SPOT), the matrix "C" becomes:
1 0 0
C= 0 cos(a) — sin(a) (25)
0 sin(a) cos(a)J
The universal observation equation for a CCD sensor becomes (Shu, 1987):
0"
Yo + XAC y (26)
42 A. BANNARI ET AL.
Up 0'
V
P = 0 (28)
.WP.
In formula (28), the coordinates x and y being null, the equations have a direc-
tion for each pixel, in other words, each pixel corresponds to a different position
of the platform (XO,YO,ZQ) and to different attitude angles (u!,(p,x)- Hence:
x =y = 0 and B =I
1
(29)
AX' = AY = AZ'=0
(translations between the systems, the instrument and the vector are null).
The universal observation equations (19) for an MSS sensor become (Shu,
1987):
xp Xo' ' 0"
YP = Yo + XAC 0 (30)
zP. .Zo_
This method has the advantage of integrating in one unique mathematical model
the data related to the satellite (attitude), the orbit, the cartographic projection,
and the image geometry by taking its centre as particular point (longitude, lat-
itude, time, viewing angle). Their simultaneous adjustment permits better de-
termination and greater robustness than step by step modelling or independent
modelling. The preprocessing of orbital data consists in calculating the osculatory
orbital parameters at the scene centre. These parameters are chosen in terms of
the orbital characteristics of the satellite and their usefulness to the cartographic
process, requiring solution for only a few seconds of orbit (Guichard, 1985). The
parameters chosen to fix an orbit in space with respect to an inertial reference
are: centre of the terrestrial reference system (O), equatorial plane (YOX), pole
of the reference plane (OZ), earth centre-to-satellite distance (p), radial velocity
of the satellite, longitude of the ascending node N (ft), inclination of the
GEOMETRIC MODELS APPLIED TO RS IMAGES 43
satellite (S)
perigee (P)
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
equatorial plane
orbital plane (/), argument of the satellite in its orbital plane (w), true anomaly
fixing the satellite on its orbit (v), intersection of the orbit plane with the refer-
ence plane (node line: AW), semi-major axis of the ellipse (a), and the orbital
eccentricity (e). Each parameter is given using a rigorous mathematical formula
(Toutin, 1986) that represents the physical realities of the viewing geometry. In
order to achieve precise rectification, the accuracy of each parameter is used to
determine which must be calculated through least squares adjustment. Figure 10
shows the geometry of the elliptical movement of the satellite in space.
The model permits the formulation of two observation equations (31) and (32)
for each control point, as opposed to the universal equations (26) and (30) where
for each scan line two equations must be expressed. It represents the collinearity
equations in a simple manner and permits translation from image coordinates to
ground coordinates, and vice versa. Due to the rigidity of the orbit and the sta-
bility of the equations, the primary precision factor is provided by the knowledge
of the nonlinear satellite attitude variations. The number and the distribution of
control points influence only slightly the quality of the results (Salamonowicz,
1986). In the case of a single image, the model permits the rigorous geometric
correction of SPOT images (Guichard, 1985; Toutin, 1991) and NOAA AVHRR
images (Moreno and Melia, 1993) with a precision of less than one pixel. In the
case of a stereoscopic SPOT image pair, if the control points are of good qual-
ity (definition and plotting) the method insures a precision in the order of one
third (1/3) of a pixel (Toutin and Carbonneau, 1989). We will provide here only
the equations and the principal characteristics, the demonstration of the model
in detail being found in the following publications: Guichard (1983 and 1985),
44 A. BANNARI ET AL.
Toutin (1985 and 1986) and Toutin et al. (1989 and 1991).
"• = 0 (31)
H
•• 0 (32)
with
where
H is the altitude of the point corrected for earth curvature,
Ho is the satellite elevation at the centre line,
No is the normal to the ellipsoid,
a is the semi-major axis of the ellipse, is a function of the rotation
of the earth,
a is the instantaneous field-of-view (rad),
X is the abscissa corrected from earth rotation and from the nonper-
pendicularity of axes,
p, q are the image coordinates,
P, Q are scale factors in Y and X, respectively,
T,6 are a function of the levelling angles in Y and X, respectively,
AT* and AR are the non-linear variations in attitude (AT*: combination of
pitch and yaw, AR: variation in roll),
x
piypihp are the ground coordinates,
b, c, x and £7 are the known parameters (2nd order), which are a function of the
geometry: satellite, scene—centre and the earth centre geometry.
Although providing a less rigorous solution than the two preceding methods
(because they do not take into consideration acquisition geometry), polynomial
methods are frequently used in digital image processing systems because the or-
bital and satellite attitude data are not always available, and because of their
simplicity. The biggest advantage of this method is the fact that all the image
distortions of the image are corrected simultaneously (Novak, 1992). Moreover,
if the acquisition geometry is stable, the earth is flat enough and if the geodetic
network is of good quality, polynomial methods provide acceptable results for
thematic mapping, but not for topographical mapping which requires high preci-
sion geometric corrections. The order of the polynomial depends on the number
of control points available. The more control information one has, the more accu-
rate the result of the rectification will be (Novak, 1992). A first order polynomial
function only permits a simple rotation of the image and does not address the
GEOMETRIC MODELS APPLIED TO RS IMAGES 45
6. CONCLUSION
Geometric correction methods are gaining importance due to the fact that digital
images must be combined with digital maps in geographic information systems.
The choice of one of the three mathematical models must be carried out in rela-
tion to the objectives of the geometric correction and precision desired. Accord-
ing to Novak (1992), the comparison of the polynomial approach and the model
based on the condition of collinearity proves that this second approach corrects
both for relief displacements and sensor distortions and yields good results, which
is in agreement with the results of Salamonowicz (1986) and Cherkaoui (1991).
The mathematical model based on the condition of collinearity related to ce-
lestial mechanics is very precise due to the fact that it integrates the parameters
of various models (attitude, sensor, earth and osculating parameters of the orbit).
The model permits the formulation of two observation equations for each control
point. The second model based solely on the condition of collinearity, and not
related to celestial mechanics, requires the formulation of two observation equa-
tions for each scan line which increases the number of equations and calculation
46 A. BANNARI ET AL.
time. This method requires very good knowledge of the image coordinates and a
more or less precise knowledge of satellite attitude. If the acquisition geometry
is stable, the earth is flat enough and if the geodetic network is of good quality,
polynomial methods provide results relatively comparable to the two preceding
models. This method may provide acceptable results for thematic mapping, but
not for topographical mapping which requires high precision geometric correc-
tions.
ACKNOWLEDGMENTS
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015
The authors would like to thank N.S.E.R.C. for their financial support (Operat-
ing Grant P 000 5252). Also, we would like to thank Mr. Paul Gagnon for his
linguistic support.
References
Bannari, A. (1989) Étude comparative du système photogrammétrique à courte distance. Thèse de
Maîtrise ès sciences, Département des sciences géodésiques et télédétection, Université Laval,
Québec, Canada, 104 pp.
Caloz, R., Blaser, T. J., and Willemin, G. (1991) Création d'une ortho-image à l'aide d'un modèle
numérique de terrain: influences des modes de rééchantillonnage radiométrique. Actes des qua-
trièmes journées scientifiques de Télédétection, PUQ/AUPELF-UREF, Montréal (Québec), Cana-
da, 17-30.
Cherkaoui, O. M. (1991) Étude comparative des images satellites et leur contribution pour la déter-
mination des points d'appui et la compilation des cartes. Thèse de Doctorat (Ph.D.), Départe-
ment des sciences géodésiques et télédétection, Université Laval, Québec, Canada, 211 pp.
CNES (1986) Guide des utilisateurs de données SPOT, Centre national d'études spatiales, Toulouse,
France, 142 pp.
Colwell, R. N. (ed.) (1983) Manual of Remote Sensing, 2nd ed. American Society of Photogrammetry,
Kills Church, VA, USA, 2440 pp.
Ghosh, K. S. (1988) Analytical Photogrammetry, 2nd ed. Pergamon Press, NY, USA, 308 pp.
Guichard, H. (1983) Étude théorique de la précision dans l'exploitation cartographique d'un satellite à
défilement: application à SPOT. Bulletin de la Société française de photogrammétrie et télédétection
90: 15-26.
Guichard, H. (1985) Etude mathématique des possibilités cartographiques de SPOT et applications
pratiques. Compte rendu du Sième congrès de l'A.Q.T., Québec, Canada, 22-24 Mai 1985, Associ-
ation Québécoise de Télédétection, Chicoutimi, 355-366.
Joly, G. (1986) Traitement des fichiers-images. Collection télédétection satellitaire (E Verger, ed.), vol.
3, Édition Paradigme, Caen, France, 178 pp.
Moik, J. G. (1980) Digital Processing of Remotely Sensed Images. Goddard Space Flight Center, Na-
tional Aeronautics and Space Administration (NASA), Washington, 330 pp.
Moreno, J. E, and Melia, J. (1993) A method for accurate geometric correction of NOAA AVHRR
HRPT data. IEEE Transactions on Geoscience and Remote Sensing 31 (1): 204-226.
Novak, K. (1992) Rectification of digital imagery. Photogrammétric Engineering and Remote Sensing
58 (3): 339-344.
Richards, J. A. (1986) Remote Sensing Digital Image Analysis. Springer-Verlag, New York, 274.
Rochon, G. (1982) Traitement numérique des images de télédétection. Département de géographie
et télédétection, Université de Sherbrooke, Sherbrooke (Québec), Canada, 7 pp.
Shu, N. (1987) Restitution géométrique des images spatiales par la méthode de l'équation de colinéar-
ité. Bulletin de la Société française de photogrammétrie et télédétection 105: 27-40.
Slama, C. C. (ed.) (1980) Manual of Photogrammetry, 4th ed., ch. 17. American Society of Photogram-
metry, Falls Church, VA, USA.
Salamonowicz, P. H. (1986) Satellite orientation and position for geometric correction of scanner
imagery. Photogrammétric Engineering and Remote Sensing 52 (4): 491-499.
GEOMETRIC MODELS APPLIED TO RS IMAGES 47
Toutin, Th. (1985) Analyse mathématique des possibilités cartographiques du système SPOT. Thèse
de Docteur ingénieur en Sciences Géodésiques à l'École Nationale des Sciences Géographiques,
Paris, France, 162 pp.
Toutin, Th. (1986) Étude mathématique pour la rectification d'images SPOT. Actes du XVIIIe Congrès
de la Fédération internationale des géomètres, Toronto, Canada, 1-11 Juin 1986, Association cana-
dienne des sciences géodésiques et cartographiques, Ottawa, 379-395.
Toutin, Th., and Carbonneau, Y. (1989) La multi-stéréoscopie pour les corrections d'images SPOT-
HRV. Journal canadien de télédétection 15 (2): 110-119.
Toutin, Th. (1991) Guide général du système de rectification d'image de télédétection. Centre cana-
dien de géomatique (ed.), Sherbrooke (Québec), Canada, 102 pp.
Downloaded by [Carnegie Mellon University] at 22:32 02 February 2015