Imaging Geometry and

Correction
GmE 202 Principles of Remote Sensing
Lecturer: Ariel C. Blanco
Introductory Digital Image Processing. 3rd edition. Jensen, 2004
Sensor Geometry
• Frame
• photogrammetric cameras; central perspective projection
• Point (across-track scanner)
• Line (along-track scanner
• Panoramic
Image Acquisition Geometry
Digital Image Processing (DIP)

• DIP involves the manipulation and interpretation of digital images

with the aid of a computer.

New
Digital Image Computer Digital Image

Data are inserted in Computation results

an equation or typically form a new
series of equations image/s
 7 Broad types of computer-assisted
operations
• Image rectification and restoration
• Correct distorted or degraded image data to create a more faithful representation of
the original scene
• Geometric distortion correction, radiometric calibration, noise removal
• Preprocessing operations
• Image enhancement
• Procedures applied to image data to more effectively display or record the data for
subsequent interpretation
• To create images from the original images to increase the amount of information
visually interpretable from the data
• Image classification
• To replace visual analysis of the image date with quantitative techniques for
automating the identification of features in a scene
• Spectral patter recognition; Spatial pattern recognition
• Data merging and GIS integration
• Hyperspectral image analysis
• Biophysical modeling
• Image transmission and compression
 Image Rectification and Restoration
(Preprocessing)
• Forms the preparatory phase for later analysis
• To correct image data for distortions or degradations that
stem from the image acquisition process
• Dependent on digital image acquisition type, platform, and
total field of view
• Typical Operations
• Geometric correction or preprocessing
• Bring image into registration with maps/images
• Radiometric correction or preprocessing
• E.g. Adjust DNs for effect of hazy atmosphere
• Noise Removal; Feature Extraction
• Preprocessing operations changes data and may introduce
artifacts not immediately observable
Geometric Distortions

• No remote sensing images are free of geometric

distortions and an essential requirement for
integrated processing of remote sensing images and
data from geographic information systems (GIS) is
that they are spatially referenced.

• The geometric distortions inherent in remote

sensing images fall into two categories:
• systematic distortions
• nonsystematic distortions.
Geometric Distortions
• Many factors contribute to the systematic distortions:
• Panoramic distortion
• Platform velocity
• Curvature of the earth
• Earth rotation
• Scan skew
• Mirror scan velocity
• This type of distortions can be rectified using data
from platform ephemeris and knowledge of internal
sensor distortion.

Ephemeris: Any tabular statement of the assigned places of a celestial

body for regular intervals
Panoramic Distortion

• Due to spacing of detectors

and regular sampling
• The ground area imaged is
proportional to the tangent
of the scan angle rather than
to the angle itself. Because
data are sampled at regular
intervals, this produces
along-scan distortion
• Increases with swath
• Corrected using orbital model
Panoramic Distortion

• Due to spacing of detectors

and regular sampling
• The ground area imaged is
proportional to the tangent
of the scan angle rather than
to the angle itself. Because
data are sampled at regular
intervals, this produces
along-scan distortion
• Increases with swath
• Corrected using orbital model
Cross-track Scan Error
• This type of error is a function of
the distance from the sensor to the
target, the instantaneous field of
view (IFOV), and the scan angle off
nadir.
• This type of error is termed
“Tangential” distortion as
compared to “Radial” distortion
present in analog aerial
photographs.
 Earth Rotation
• Due to time taken to build an image
as the sensor scans the earth
surface features
• Each line offset to the west from
the previous one
• Corrected using orbital model
Platform Velocity

• If the speed of the platform changes, the ground track covered by

successive mirror scans changes, producing along-track scale
distortion
 Scan Skew
• Caused by the forward motion if the platform during the time required
for each mirror sweep. The ground swath is not normal to the ground
track but is slightly skewed, producing cross-scan geometric distortion

Mirror Scan Velocity Variance

 The mirror scanning rate is usually not
constant across a given scan, producing
along-scan geometric distortion.
Geometric Distortions
•Nonsystematic distortions arise from
• Sensor system’s attitude, velocity, and
altitude
• Can be corrected only through the use of
ground control points (GCPs).
• In particular, topographic, or relief
displacement due to terrain variation is
usually the most serious of the displacement
types, especially in mountainous terrain.
Effect of Platform Position and Attitude Errors
Effect of Platform Position and
Attitude Errors
Terrain-related Distortions

• Due to small changes in altitude and aspect

• Can be corrected by orthorectification
• Requires a digital elevation model (DEM)
• Can be corrected using a “rubbersheet” rectification based on ground
control points.
Image Rectification
• There are essentially two different categories of image
rectification techniques. One is the deterministic
approach that establishes models for the nature and
magnitude of the sources of distortion and uses these
models to establish correction formulae.
• The deterministic approach relies on data of the flight
parameters and the terrain information, and is
effective when the types of distortion are well
characterized, such as that caused by earth rotation.
Image Rectification
• The other approach of image rectification is the statistical
approach, which, by means of a GCP data set, establishes
mathematical relationship between image coordinates and their
corresponding map coordinates using standard statistical
procedures.
• These relationships can be used to correct the image geometry
irrespective of the analyst’s knowledge of the source and type of
distortion.
• The most widely used method in this category is the polynomial trend
mapping (PTM) technique that employs polynomial regression equations
to relate image coordinates and their corresponding map coordinates.
 Image Rectification
• The order of polynomials to be used depends on the
image sources and the terrain variation in the scene.
• For airborne images or images from areas of rugged terrain,
more severe distortions present and higher order polynomials
are used.
• In contrast, affine transformation, a first-order
polynomial, may be sufficient to rectify satellite images.
• An extended model of the conventional polynomial mapping
is the Hardy’s multiquadric interpolation function (MIF), which
yields zero errors for coordinate mapping at GCPs (Hardy,
1990, Star et al., 1997). The MIF method employs a distance-
weighting algorithm to adjust residuals from the PTM method.
Rectification, Georeferencing

• Georeferencing refers to assigning map coordinates

to image data. The image data may be projected
onto the desired plane, but not yet referenced to
the proper coordinate system.
• Rectification, by definition, involves georeferencing,
since all map projection systems are associated with
map coordinates.
• Image-to-image registration involves georeferencing
only if the reference image is already
georeferenced.
• Georeferencing, by itself, may involve changing only
the map coordinate information in the image file.
 Orthorectification
• Orthorectification is a form of rectification that corrects for terrain or relief
displacement and can be used if there is a DEM or DSM of the study area. It is
based on collinearity equations, which can be derived by using 3D GCPs. In
relatively flat areas, orthorectification is not necessary, but in mountainous areas
(or on aerial photographs of buildings), where a high degree of accuracy is
required, orthorectification is recommended.
Map Projection and Coordinate
System
• Before rectifying the data, you must determine the
appropriate coordinate system considering the
primary use for the data.
• How large or small an area is mapped?
• Different projections are intended for different size areas.
• Where on the globe is the study area?
• Polar regions and equatorial regions require different
projections for maximum accuracy.
• What is the extent of the study area?
• Circular, north-south, east-west, and oblique areas may all
require different projection systems.
 Universal Transverse Mercator
Coordinate System
• Transverse Mercator projection
• The UTM system divides the surface of Earth between
80°S latitude and 84°N latitude into 60 zones, each 6°
of longitude in width and centered over a meridian of
longitude. Zones are numbered from 1 to 60.
• Reference longitude: Central meridian of each zone
• Reference latitude is the Equator
• X-shift, Y-shift; False easting and northing
Universal Transverse Mercator Grid
When to Rectify

• Rectification is necessary in cases where the pixel grid of

the image must be changed to fit a map projection system
or a reference image. There are several reasons for
rectifying image data:
• comparing pixels scene to scene in applications, such as change
detection
• developing GIS data bases for GIS modeling
• identifying training samples according to map coordinates prior to
classification
• creating accurate scaled photomaps
• overlaying an image with vector data
• comparing images that are originally at different scales
• extracting accurate distance and area measurements
• mosaicking images
• performing any other analyses requiring precise geographic locations
Georeference

• Planar images, that need map coordinate

information should only be georeferenced, (a much
simpler process).
• In many cases, the image header can simply be
updated with new map coordinate information.
• Redefining:
• the map coordinate of the upper left corner of the image
• the cell size (the area represented by each pixel)
• This information is usually the same for each layer of an image
file, although it could be different. For example, the cell size of
band 6 of Landsat TM data is different than the cell size of the
other bands.
Rectification

• Use methods based on statistical operations: “Ground

Control Point rectification” (Star & Estes 1990).
• Localization of visible points in the images are related to
same points in reality (or on maps).
• Establish relationship between the image coordinate system
and the map coordinates
• Together they form polynoms and their coefficients
calculated by regression (Star et al 1990).
• Error is given as RMS error (Root Mean Square) that denote
difference between output location for a GCP and the real
coordinates for the same point when the point is recalculated
via a matrix of transformation.
• Use interpolation to determine values for the new grid point.
Ground Control Points

• GCPs are specific pixels in an image for which the output map
coordinates (or other output coordinates) are known. GCPs
consist of two X,Y pairs of coordinates:
• Source coordinates—usually data file coordinates in the
image being rectified
• Reference coordinates—the coordinates of the map or
reference image to which the source image is being
registered
• The term map coordinates is sometimes used loosely to apply to
reference coordinates and rectified coordinates. These coordinates
are not limited to map coordinates.
• For example, in image-to-image registration, map coordinates are not
necessary.
GCP’s
GCP’s
Distribution of GCP’s
GCP Selection for High Resolution Satellite Image
 Two-Dimensional
Geographic Transformations

Y0 Yn Yn

Yn
Y0
X0
Xn
Xn
Xn X0

Translation Scaling Rotation

Xn = sX ×X0 Xn = X0 ·cos() + Y0 ·sin()
Xn = X0 ± DX0
Yn = Y0 Yn = Yn = - X0 ·sin() + Y0 ·cos()
± DY0 sY × Y0
 Conformal Transformation
• Moves and rotates objects in 2D and 3D space. Additionally, you can
scale the objects based on alignment points when using the 2D option.
 Isogonal Affine Transformation
or Conformal/Similarity Transformation
• Isogonal: having equal angles
• Impose additional condition of equal scale (S = Cx = Cy)
yielding 4 parameters: S, , DX0, DY0
• Moves and rotates and scale objects in 2D space.

 Xn   a b   Xo   c   cos  sin    Xo   DXo 

 Yn    b a  *  Yo    d   S *   sin  *   
cos    Yo   DYo 

        

4 parameters: S –scale, α rotation, DX0, DY0 shifts in X and Y.

Xn,Yn are the transformed coordinates.
Xo, Yo are the original coordinates.
Two given points are required ( X1,Y1 and X2,Y2)
 Conformal Transformation
• Moves and rotates and scales objects in 2D space.
• Often called similarity Transformation since the basic shape
remain similar after the transformation
 Conformal Transformation

• The formulas can be written in different forms

1. To compute the parameters given the coordinates
2. To compute the new coordinates given the parameters

Xn  S  cos   X 0  S  sin   Y0  DX 0
Yn   S  sin     X 0  S  cos     Y0  DY0

a 
Xn  a  X 0  b  Y0  c  Xn   X 0 Y0 1 0  b 
 Yn    Y   
X0 0 1  c 
Yn  b  X 0  a  Y0  d    0
 
d 
 Conformal/Similarity Transformation
• The formulas can be written in different forms
3. To compute the old coordinates given the parameters and new
coordinates (back substitution)

Xn  a  X 0  b  Y0  c ( Xn  c)  a  X 0  b  Y0
Yn  b  X 0  a  Y0  d (Yn  d )  b  X 0  a  Y0
Add equations to get X0
Multiply Eq. 1 by a, Eq 2 by -b
a  Xn  c   b  Yn  d 
X0 
a ( Xn  c )  a  X 0  ab  Y0
2
a2  b2

b(Yn  d )  b 2  X 0  ab  Y0 b  Xn  c   a  Yn  d 
Similar operation to obtain Y0: Y0 
a 2  b2
Multiply Eq. 1 by b, Eq 2 by a
 Conformal Transformation: Example
• Moves, rotates and scale objects in 2D space.

No Xo Yo Xn Yn
(map) (map) (ground) (ground)
1 10 10 350 190
2 80 60 250 300

 Xn   a b   Xo  Cx 
 Yn     b a  *  Yo   Cy 
       
350  a  10  b  10  C x 350 10 10 1 0  a 
190  b  10  a  10  Cy
  10  10
 190 0 1  b 
  * 
250  a  80  b  60  C x 250 80 60 1 0  C x 
300  b  80  a  60  C y 300 60  80
   0 1 C y 
 Conformal Transformation: Example
 a  10 10 1 0 1 350   0.203 
 b  10  10 0 1 190    1.716 
   *    
C x  80 60 1 0 250 369.189
C  60  80 0 1 300 174.865 
 y     

• Transform a given point X0=121.48, Y0=22.78

 Xn    0.203  1.716 121.48 369.189  310.59 

 Yn    1.716  0.203 *  22.78   174.865   373.6773
         

• The point is transformed to Xn=310.59, Yn=373.6773

 Affine Transformation
• Used in photogrammetry for:
• Transform comparator coordinates to photo coordinates and
used for correcting film distortion
• Transform model coordinates to survey coordinates
• Property
• Carry parallel lines y 2

into parallel lines 

• Does not have to 

preserve orthogonality 2t
t1 y1
 Yn 

Affine Transformation 

DX DY
Physical interpretation: Xn

Xn  C x   X 0   cos( )  C y   Y0   sin       DX 0
Yn  Cx   X 0   sin     C y   Y0   cos       DY0
6 parameters: Cx, Cy, , , Dx0, Dy0, and in linear form:
a  Cx cos  d  C x sin   
b  C y sin      e  C y cos     
c  Dx0 f  Dy0

• 
𝑋𝑛 𝑎 𝑏 𝑋𝑜 𝑐
[ ][ 𝑌𝑛
=
𝑐 𝑑
∗
] [ ][]
𝑌𝑜
+
𝑓
 2-D Affine Transformation
• The formulas for an affine transformation:
 xT   a A bA   xs   c A 
 y   d      
e A   ys   f A 
 T  A
• If n control points are measured, this Equation is reorganized as
follows: a  A
 xT 1   xs1 ys1 1 0 0 0  
y   0   bA 
 T1   0 0 xs1 ys1 1
 cA 
          
    d A 
 xTn   xsn ysn 1 0 0 0
 eA 
 yTn   0 0 0 xsn ysn 1   
             f A 
y A 

 Orthogonal Affine Transformation
• Impose condition of orthogonality ( = 0) yielding 5
parameters: Cx, Cy, , x’, y’

Xn  Cx   X 0   cos   C y   Y0   sin   DX 0
Yn  C x   X 0   sin     C y   Y0   cos     DY0
 Rigid Body Transformation

• Condition: orthogonality and no scale change (Cx =

Cy = 1)

Xn  X 0  cos   Y0  sin   DX 0
Yn   X 0  sin   Y0  cos   DY0
• 3 parameters: , x’, y’
 Projective or Polynomial Transformation
• Instead of 4 or 6 parameters we have many parameters at
least 8 (8 would be the Bi-linear or projective
transformation)
• With more parameters we need more known points to solve
the equations
• N-equations and N unknowns.

Xn  a0  a1 X 0  a2Y0  a3Y0 X 0  a4 X 0 2  a5Y0 2  ...

2 2
Yn  b0  b1 X 0  b2Y0  b3Y0 X 0  b4 X 0  b5Y0  ...
Transformations
Polynomial Equations

• Depending upon the distortion in the imagery, the

number of GCPs used, and their locations relative to
one another, complex polynomial equations may be
required to express the needed transformation.

• The degree of complexity of the polynomial is expressed

as the order of the polynomial. The order is simply the
highest exponent used in the polynomial.
• The order of transformation is the order of the
polynomial used in the transformation.
• Usually, 1st-order or 2nd-order transformations are used.
1st-Order or 2nd-Order

• Most rectifications are either 1st-order or 2nd-order. The danger of

using higher order rectifications is that the more complicated the
equation for the transformation, the less regular and predictable the
results are.
• To fit all of the GCPs, there may be very high distortion in the
image.
Polynomial Curves vs. GCPs

• Every GCP influences the coefficients, even if there is not a

perfect fit of each GCP to the polynomial that the
coefficients represent. The distance between the GCP
reference coordinate and the curve is called RMS error. The
least squares regression method is used to calculate the
transformation matrix from the GCPs.
Linear Transformations:
• A 1st-order transformation is a linear transformation.
• It can change:
• location in X and/or Y
• scale in X and/or Y
• skew in X and/or Y
• rotation
First-order Transformation

• First-order transformations can be used to project raw

imagery to a planar map projection, convert a planar map
projection to another planar map projection, and when
rectifying relatively small image areas.
• Linear transformations may be required before collecting
GCPs on the displayed image.
• A 1st-order transformation can also be used for data that are
already projected onto a plane, but not rectified to the
desired map projection.
• When doing this type of rectification, it is not advisable to
increase the order of transformation if at first a high RMS
error occurs.
• Examine other factors first, such as the GCP source and distribution,
and look for systematic errors.
Residuals and RMS Error per
GCP
• Residuals are the distances between the source and retransformed
coordinates in one direction. They are shown for each GCP.
• If the GCPs are consistently off in either the X or the Y direction, more
points should be added in that direction.
• This is a common problem in off-nadir data.
 RMS Error

RMS error is the distance between the

input (source) location of a GCP and the
retransformed location for the same GCP.
(the difference between the desired output
coordinate for a GCP and the actual output
coordinate for the same point, when the
point is transformed).
RMS Error Tolerance

• In most cases, it is
advantageous to tolerate a
certain amount of error
rather than take a more
complex transformation.
• The amount of RMS error
that is tolerated can be
thought of as a window For example, if the RMS error
around each source tolerance is 2, then the
coordinate, inside which a retransformed pixel can be 2 pixels
retransformed coordinate is away from the source pixel and still
considered to be correct be considered accurate.
(that is, close enough to use).
RMS Error Tolerance

• Acceptable RMS error is determined by the end use, the

type of data being used, and the accuracy of the GCPs and
ancillary data being used.
• For example, GCPs acquired from GPS should have an
accuracy of about 10 m, but GCPs from 1:24,000-scale maps
should have an accuracy of about 20 m.
• It is important to remember that RMS error is reported in
pixels.
• Therefore, if you are rectifying Landsat TM data and want the
rectification to be accurate to within 30 meters, the RMS error should
not exceed 1.00.
• Acceptable accuracy depends on the image area and the
particular project.
Evaluating RMSE

• To determine the order of polynomial transformation, you can assess

the relative distortion in going from image to map or map to map.
• One should start with a 1st-order transformation unless it is known
that it does not work. It is possible to repeatedly compute
transformation matrices until an acceptable RMS error is reached.
 Options?
• Throw out the GCP with the highest RMS error, assuming that
this GCP is the least accurate. Another transformation can then
be computed from the remaining GCPs. A closer fit should be
possible. However, if this is the only GCP in a particular region of
the image, it may cause greater error to remove it.
• Tolerate a higher amount of RMS error.
• Increase the complexity of transformation, creating more
complex geometric alterations in the image. A transformation
can then be computed that can accommodate the GCPs with less
error.
• Select only the points for which you have the most confidence.

The next step in the rectification/registration process is to create the output file. Since
the grid of pixels in the source image rarely matches the grid for the reference image,
the pixels are resampled so that new data file values for the output file can be
calculated.
Second-order (and higher)
Transformations
• Transformations of the 2nd-order or higher are nonlinear
transformations that can correct nonlinear distortions. The
process of correcting nonlinear distortions is also known as
rubber sheeting.
• Second-order transformations can be used to convert
Lat/Lon data to a planar projection, for data covering a large
area (to account for the Earth’s curvature), and with
distorted data (for example, due to camera lens distortion).
• Third-order transformations are used with distorted aerial
photographs, on scans of warped maps and with radar
imagery.
• Fourth-order transformations can be used on very distorted
aerial photographs.
ts of Order

• The computation and output of a

higher-order polynomial equation
are more complex and are used to
perform more complicated image
rectifications.
• The examples use only one
coordinate (X), instead of two
(X,Y), which are used in the
polynomials for rectification.
• This enables you to draw two-
dimensional graphs that illustrate the
way that higher orders of
transformation affect the output
image.
“Rubber-sheet”

• In ”rubber-sheet” transformation, the image is stretched

to fit most of the coordinates. Of importance is to have
many GCPs with good coverage all over the image.
• Rubbersheet process builds a numeric coordinate
transformation between the original image coordinates
and the rectified map. The method is nonlinear.
• For image rectification, the known control points can be
triangulated into many triangles. Each triangle has three
control points as its vertices.
• Then, the polynomial transformation can be used to establish
mathematical relationships between source and destination
systems for each triangle.
Rubber-sheeting

• Because the transformation exactly passes through

each control point and is not in a uniform manner,
finite element analysis is also called rubber sheeting.
It can also be called as the triangle-based rectification
because the transformation and resampling for image
rectification are performed on a triangle-by-triangle
basis.
• This triangle-based technique should be used when
other rectification methods such as polynomial
transformation and photogrammetric modeling
cannot produce acceptable results.
Check Point Analysis

• For an exact modeling method like rubber sheeting,

the ground control points, which are used in the
modeling process, do not have much geometric
residuals remaining.
• Independent check point analysis is critical for
determining the accuracy of rubber sheeting
modeling.
• To evaluate the geometric transformation between
source and destination coordinate systems, the
accuracy assessment using independent check points
is recommended.
Hold-out validation (HOV)

• Most used method

• Consists of partitioning the known ground points in two
sets, the first used into the orientation-
orthorectification model (GCPs – Ground Control
Points) and the second to validate the model itself (CPs
– Check Points)
• The accuracy is just the RMSE of residuals between
imagery derived coordinates with respect to CPs
coordinates.
• Generally not reliable and it is not applicable when a
low number of ground points is available
Leave-one-out cross-validation
(LOOCV)

• Iterative application of the orthorectification model using all

the known ground points (or a subset of them) as GCPs
except one, different in each iteration, used as CP.
• In every iteration the residual between imagery derived
coordinates with respect to CP coordinates (prediction error
of the model on CP coordinates) is calculated;
• The overall spatial accuracy achievable from the
orthorectified image may be estimated by calculating the
usual RMSE or, better, a robust accuracy index like the mAD
(median Absolute Deviation) of the prediction errors on all
the iterations.
Other Forms of
Accuracy Assessment
Interpolation of Reflectance Values
Resampling

• The process of extrapolating data

values to a new grid. Resampling is
the step in rectifying an image that
calculates pixel values for the
rectified grid from the original data
grid.
• Images treated as simple array to be
manipulated to create another array
• Input image represented by open
dots
• Output image represented by solid
dots
• Location of output pixels
determined from locational
information from GCPs
• Establish geometry of output and
relationship with Input
 Resampling

The input image with source GCPs

The output grid, with reference GCPs shown.

To compare the two grids, the input Using a resampling method, the pixel
image is laid over the output grid, so that values of the input image are assigned
the GCPs of the two grids fit together. to pixels in the output grid.
Nearest Neighbor

• The nearest neighbor approach

uses the value of the closest input
pixel for the output pixel value.
• To determine the nearest neighbor,
the algorithm uses the inverse of
the transformation matrix to
calculate the image file coordinates
of the desired geographic
coordinate. The pixel value
occupying the closest image file
coordinate to the estimated
coordinate will be used for the
output pixel value in the
georeferenced image.
Nearest Neighbor
ADVANTAGES:
• Output values are the original input values. Other methods of resampling
tend to average surrounding values. This may be an important
consideration when discriminating between vegetation types or locating
boundaries.
• Since original data are retained, this method is recommended before
classification.
• Easy to compute and therefore fastest to use.
DISADVANTAGES:
• Produces a choppy, "stair-stepped" effect. The image has a rough
appearance relative to the original unrectified data.
• Data values may be lost, while other values may be duplicated.
Bilinear Interpolation

• Weighted average of four

nearest neighbors
• Adv: Image has more
“natural” look
• Brightness values of original
scene lost
• Decreased resolution by
averaging over areas – gives
kind of smearing effect
• Is computationally more
expensive than nearest
neighbor
Bilinear Interpolation
Cubic Convolution

••Uses
Uses weighted
weighted average
average
of
of values
values of
of 16
16 pixels
pixels
••Images
Images generally
generally more
more
attractive
attractive than
than other
other
procedures
procedures
••Data
Data altered
altered more
more
••Computations
Computations moremore
intensive
intensive and
and number
number of
of
GCPs
GCPs is
is higher
higher
Original image Nearest neighbor

Bilinear Interpolation Cubic Convolution

Disadvantages of Rectification

• During rectification, the data file values of rectified

pixels must be resampled to fit into a new grid of pixel
rows and columns. Although some of the algorithms
for calculating these values are highly reliable, some
spectral integrity of the data can be lost during
rectification.
• If map coordinates or map units are not needed in the
application, then it may be wiser not to rectify the
image.
• An unrectified image is more spectrally correct than a
rectified image.
Pre-classification?

• Some analysts recommend classification before

rectification, since the classification is then based
on the original data values.
• Another benefit is that a thematic file has only one
band to rectify instead of the multiple bands of a
continuous file.
• On the other hand, it may be beneficial to rectify
the data first, especially when using GPS data for
the GCPs. Since these data are very accurate, the
classification may be more accurate if the new
coordinates help to locate better training samples.
 Subset

• Portion of larger images showing only

region of interest
• Complexity
• Subsets must be “registered” or matched to
other data – need distinctive landmarks in
both cases
• More convenient to prepare prior to
registration
• Time & computational effort increase with
larger images
• Convenient to prepare preliminary
subset before selecting final, smaller
set