You are on page 1of 50

I ntroduction to Geometric

Correction
Week 3 Geometric Correction Sept. 20-24, 1999
Radiometric and Geometric Correction
Correcting for errors in the geometry and
measured brightness values of pixels on
digital images
Week 3 Geometric Correction Sept. 20-24, 1999
Sources of Radiometric Distortion
errors related to measured brightness values
of the pixels which result from the
instruments that are used to record the data

errors resulting from effects of the
atmosphere
Week 3 Geometric Correction Sept. 20-24, 1999
I nstrumentation Errors
IDEAL - radiation detector should have a
proportional increase and decrease of signal
with detected radiation level
REAL - many have small degree of non-
linearity and will also give small signal out
even when no radiation is being detected.
Each detector will have slightly different
transfer characteristics.
Leads to striping on an image
Week 3 Geometric Correction Sept. 20-24, 1999
I nstrumentation Errors
Gain
1
2
3
4
5
6
Signal
out
Radiation in
Offset
IDEAL REAL Signal
out
Offset
Radiation in
Bands
(Richards, 1993)
Week 3 Geometric Correction Sept. 20-24, 1999
Atmospheric Errors
Wide field of view of sensor = difference in
atmospheric path length between nadir and
extremities of the swath
Rayleigh and Mie scattering will effect
different wavelength bands of a sensor
system.
Week 3 Geometric Correction Sept. 20-24, 1999
Sources of Geometric Distortion
Rotation of the earth during image acquisition
the finite scan rate of some sensors
the curvature of the earth
variation in platform altitude, attitude and velocity
panoramic effects related to the imaging geometry
Errors of geometry in the digital image:
Week 3 Geometric Correction Sept. 20-24, 1999
Rotation of the Earth
Takes a finite amount of time to obtain a
frame of image data
earth is rotating from west to east, therefore
the bottom of the image must be off-set the
ground distance that the earth rotated during
the image acquisition
Week 3 Geometric Correction Sept. 20-24, 1999
Effect of Rotation of the Earth
Satellite
motion
Earth Motion
Week 3 Geometric Correction Sept. 20-24, 1999
Finite Scan Rate of Sensors
Mechanical line scanners take a finite
amount of time to scan one line across the
swath
satellite is moving in its orbit during the
scan time
end of scan line has advanced a specific
distance as compared to its beginning
Week 3 Geometric Correction Sept. 20-24, 1999
Curvature of the Earth
Aircraft due to their low altitude are not effected
Landsat and SPOT are not effected because they
have relatively narrow swaths
NOAA has swath of 2700 km and pixels at the
edges of the swath would be different sizes
Week 3 Geometric Correction Sept. 20-24, 1999
Curvature of the Earth
Satellite
Swath
Earths surface
(Richards, 1993)
Week 3 Geometric Correction Sept. 20-24, 1999
Variation in Altitude, Velocity
and Attitude
Changes in the sensors altitude result in
changes in scale in the IFOV
velocity results in change of scale along the
track direction
attitude (pitch, roll, yaw) of aircraft result in
localized distortions
Week 3 Geometric Correction Sept. 20-24, 1999
Panoramic Distortion
Pixel size on the ground is larger at the
extremities of the scan than at nadir
along track, pixels are same
Week 3 Geometric Correction Sept. 20-24, 1999
Types of Distortion in an I mage
Systematic
predictable
corrected at the receiving station

Random
unpredictable
corrected by the analyst
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Systematic Errors
Caused by the eastward rotation of the earth
de-skewing corrects for this by shifting scan
lines by a calculated number of pixels
dependent on the estimated latitude relative to
the start point, satellite velocity relative to the
Earth, and the length of the image frame being
recorded

reason for the parallelogram appearance
Week 3 Geometric Correction Sept. 20-24, 1999
Parallelogram I mage Shape
Original SPOT image of the Jakarta area
Bogor
study area
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Systematic Errors
Other errors are dependent upon the type
of sensor
Week 3 Geometric Correction Sept. 20-24, 1999
Non-Systematic Errors
Altitude and attitude variations
if the sensor changes from its normal orbit,
distortions in scale result

topographic effects
cause distortions in scale - orthorectification

no relation of raw image to coordinate
system

Week 3 Geometric Correction Sept. 20-24, 1999
Non-Systematic Errors
noise
random or irregular interfering effects in the
data which effect its quality
SENSOR- destriping corrects for problems
associated with sensor mirror effects, line-dropout
correction involves correcting for dropped lines
HAZE - remove atmospheric scatter
Geometric Correction
Correcting for errors of image
geometry
Week 3 Geometric Correction Sept. 20-24, 1999
Raw Data For Bogor
No defined
datum or
map
projection.
Still
oriented to
the
satellites
orbit
Week 3 Geometric Correction Sept. 20-24, 1999
Ground Control Point (GCP)
Collection
User identifies x,y coordinates of several
pairs of points (GCPs) that represent the
same area within both the uncorrected and
corrected data.
Uncorrected (Bogor image) is in units of
pixels and lines, while the corrected road
vectors recorded with coordinates of
eastings and northings or latitude and
longitude.
Week 3 Geometric Correction Sept. 20-24, 1999
ER Mapper Rectification Options
Or Triangulated
Or quadratic, cubic
Or bilinear, cubic
Week 3 Geometric Correction Sept. 20-24, 1999
Type of Rectification
Polynomial
transform raw (or unknown) imagery to a
known projection.
Georeferencing, geocoding
use GCPs
reduces global image distortion

Triangulation
reduces local distortion in the image
common in geo-scanned data.
Delauney Triangle is a common method
Week 3 Geometric Correction Sept. 20-24, 1999
1. Collect GCPs
source from vectors, digital maps, GPS, hardcopy maps,etc
even distribution

2. Solve polynomial equation
relates old raw image to new corrected image
software does this

3. Transformation
Use equation solution to transform the entire image

4. Resampling
Interpret spatially which pixel goes where in the new
coordinate system
I mage Correction Steps
Week 3 Geometric Correction Sept. 20-24, 1999
Polynomial Rectification
1st Order Coordinate Transformation
x = a
0
+ a
1
x + a
2
y
y = b
0
+ b
1
x + b
2
y

where: x and y = output (correct map coordinates)
x and y = input (distorted image coordinates)

These are linear, 1st order equations with 3
unknown coefficients.
Week 3 Geometric Correction Sept. 20-24, 1999
To solve the 1st order equations:

need 3 GCPs - one equation for each unknown

x = a
0
+ a
1
x + a
2
y
x
1
= a
0
+ a
1
x
1
+ a
2
y
1

x
2
= a
0
+ a
1
x
2
+ a
2
y
2

1 GCP - translation
2 GCPs - translation and scaling (scaling only if
non-linear)
3 GCPs - translation, scaling, and rotation - 1st
order transformation

Week 3 Geometric Correction Sept. 20-24, 1999
# GCPs for dataset : C:\Data\RS\correction_spot\bogor.ers
#
# Total number of GCPs: 15
# Number turned on : 15
# Warp order : 0
# GCP TO map projection details:
# Map Projection : SUTM48
# Datum : IND74
# Rotation : 0.000
#
# Point On Locked Cell-X Cell-Y To-X To-Y
"1" Yes Yes 813.967 742.184 702561.0928000 9270507.9415000
"2" Yes Yes 863.651 575.005 704057.8375000 9273675.8383000
"3" Yes Yes 962.961 210.023 707125.7190000 9280593.8346000
"4" Yes Yes 969.174 812.828 705401.4677000 9268664.7627000
"5" Yes Yes 1039.732 845.143 706696.0082000 9267816.1122000
"6" Yes Yes 687.160 808.980 699822.2845000 9269554.5992000
"7" Yes Yes 409.441 927.836 693970.5613000 9268035.4819000
"8" Yes Yes 418.656 307.535 696053.7386000 9280292.3662000
"9" Yes Yes 593.043 325.145 699462.0410000 9279421.5655000
"10" Yes Yes 377.656 605.094 694333.5672000 9274527.7801000
"11" Yes Yes 854.482 420.092 704327.3229000 9276774.7082000
"12" Yes Yes 637.379 537.350 699691.1522000 9275084.0113000
"13" Yes Yes 812.502 303.407 703859.1500000 9279208.6238000
"14" Yes Yes 577.939 774.541 697764.3307000 9270593.4779000
"15" Yes Yes 842.336 864.246 702731.9911000 9268026.5549000
Sample of ERMapper GCP File
Week 3 Geometric Correction Sept. 20-24, 1999
Lets take an example
Point Cell-X Cell-Y To-X To-Y
"1" 813.967 742.184 702561.0928000 9270507.9415000
"2" 863.651 575.005 704057.8375000 9273675.8383000
"3" 962.961 210.023 707125.7190000 9280593.8346000
These are the first three GCPs from Rosies
raw_bogor.gcp file.
Week 3 Geometric Correction Sept. 20-24, 1999
Linear Solution
Solution from solving for three unknown
variables (done by the computer):

x = 23796.827 + 0.0457859x - 0.005949 y
Now each additional GCP will slightly
modify these coefficients to get a better
model.
For transformation, the computer will
use the final equation and apply it to each
pixel to determine output.
Week 3 Geometric Correction Sept. 20-24, 1999
Root Mean Squared Error

requires at least 4 GCPs

compares the predicted location of the GCP in
the corrected coordinate space to the actual
location of the GCP as chosen by the operator

RMS error = ((x - x
orig
)
2
+ (y - y
orig
)
2
)



Evaluating Error
Week 3 Geometric Correction Sept. 20-24, 1999

# RMS error report:
# -----ACTUAL----- ---POLYNOMIAL---
# Point Cell-X Cell-Y Cell-X Cell-Y RMS
# "1" 813.967 742.184 814.750 742.544 0.8616
# "2" 863.651 575.005 864.378 574.977 0.7278
# "3" 962.961 210.023 962.847 210.532 0.5214
# "4" 969.174 812.828 969.125 812.585 0.2483
# "5" 1039.732 845.143 1039.550 844.931 0.2792
# "6" 687.160 808.980 686.821 809.900 0.9808
# "7" 409.441 927.836 409.521 928.235 0.4063
# "8" 418.656 307.535 418.452 307.333 0.2871
# "9" 593.043 325.145 593.418 325.136 0.3756
# "10" 377.656 605.094 377.692 604.829 0.2668
# "11" 854.482 420.092 853.937 419.899 0.5785
# "12" 637.379 537.350 637.972 537.717 0.6976
# "13" 812.502 303.407 812.168 303.127 0.4363
# "14" 577.939 774.541 577.244 773.804 1.0132
# "15" 842.336 864.246 842.205 863.861 0.4068
#
# Average RMS error : 0.539
# Total RMS error : 8.087
# End of GCP details



RMS Error
Week 3 Geometric Correction Sept. 20-24, 1999
Example of RMS Error
# RMS error report:
# -----ACTUAL----- ---POLYNOMIAL---
#Point Cell-X Cell-Y Cell-X Cell-Y RMS
# "1" 813.967 742.184 814.750 742.544 0.8616
RMS
error
= ((X-X
orig
)
2
+ ((Y-Y
orig
)
2
)
1/2

RMS
error
= ((814.750-813.967)
2
+ ((742.544-742.184)
2
)
1/2

RMS
error
= 0.862
Week 3 Geometric Correction Sept. 20-24, 1999
Second Order:
y
1
= a
0
+a
1
x +a
2
y +a
3
xy + a
4
x
2
+ a
5
y
2
Requires 6 GCPs to solve.
Requires 7 GCPs for RMS.


Third Order:
y
1
= a
0
+a
1
x +a
2
y +a
3
xy + a
4
x
2
+ a
5
y
2
+ a
6
x
2
y +a
7
xy
2
+a
8
x
3
+a
9
y
3

Requires 10 GCPs to solve, 11 for RMS.

Higher Orders
Week 3 Geometric Correction Sept. 20-24, 1999
GCP Recommendations
Always collect at least twice as many as are
needed for the polynomial order you choose
6 for 1st order
12 for 2nd order
20 for 3rd order
Keep your average RMS error below 1
pixel.
Week 3 Geometric Correction Sept. 20-24, 1999
interpolation technique to determine which pixel
from the uncorrected imagery goes where in the
corrected imagery
mismatch from transformation
3 Types:
nearest neighbor
bilinear
cubic convolution
Resampling
Week 3 Geometric Correction Sept. 20-24, 1999

DN value in the corrected grid determined from the DN
value of the pixel nearest to it in the uncorrected image

computationally efficient

does not alter the original input pixel value

features can be offset by a magnitude of a pixel

resulting in a blocky or step-like appearance

perfect for thematic or classified imagery, since classes
are not modified
Nearest Neighbor Resampling
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Nearest Neighbour
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Nearest Neighbor
Above: before resampling
Below: after resampling
Week 3 Geometric Correction Sept. 20-24, 1999
assigns output pixel values by interpolating brightness
values in two orthogonal directions in the input image

weighted average technique

more computationally intensive than nearest neighbor

should not be used for thematic or classified data

loss in image resolution due to smoothing or blurring
effect
example - Richards
Bilinear Resampling
Week 3 Geometric Correction Sept. 20-24, 1999
Bilinear I nterpolation Example
Uses 3 interpolations over the 4 closest pixels
that surround the point found on the image

Performs a weighted average using a 2x2 array
which is based on distance
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Bilinear I nterpolation
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Bilinear I nterpolation
Above: results of nearest
neighbour resampling
Below: results of bilinear
interpolation resampling
Week 3 Geometric Correction Sept. 20-24, 1999
output DNs are assigned on the basis of a weighted
average of input DNs from the 16 surrounding pixels

much less blurring than bilinear interpolation

higher computational cost

cubic polynomial fitted along the four lines of four
pixels surrounding the point in the image to form four
interpolants

5
th
order polynomial fitted through these

Cubic Convolution Resampling
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Cubic Convolution
Week 3 Geometric Correction Sept. 20-24, 1999
Example of Cubic Convolution
Above: example of bilinear
interpolation
Below: example of cubic
convolution
Week 3 Geometric Correction Sept. 20-24, 1999
Summary of Resampling Methods
Nearest neighbour - simple to compute,
grey level unaltered. Disadvantages: image
distorted, up to half a pixel offset
Bilinear Interpolation - smooth image,
geometrically accurate. Disadvantages:
values are altered, possible blurring of the
image, more computer intensive than
nearest neighbour
Week 3 Geometric Correction Sept. 20-24, 1999
Summary of Resampling Methods
Cubic Convolution - very smooth image.
Disadvantages: DN values are altered, most
computer intensive of the three sampling
methods described
Week 3 Geometric Correction Sept. 20-24, 1999
Geometrically Corrected I mage
of Bogor
Corrected Image
of Bogor - note
that it is no longer
oriented to the
path of the
satellite and has
been rotated
approx. 8. It
now has a defined
datum and map
projection.

You might also like