Digital Imaging and Remote Sensing Laboratory
Correction of Geometric Distortions in Correction of Geometric Distortions in Line Scanner ImageryLine Scanner Imagery
Peter Kopacz
Dr. John Schott
Bryce Nordgren
Scott Brown
May 8, 1998
Digital Imaging and Remote Sensing Laboratory
CoverageCoverage
• Line Scanner Background Theory
• Geometric Distortions
• Results
• Future Work / Recommendations
Digital Imaging and Remote Sensing Laboratory
Line Scanner ExampleLine Scanner Example
across track
along track
FOV
IFOV
sampled ground pixel
• Airborne / Spaceborne EO Imaging Instrument
• Scanning mirror allows for collection of ground data
one line at a timegeometric distortions...
Digital Imaging and Remote Sensing Laboratory
Geometric DistortionsGeometric Distortions
• Degradation in visual appearance of the
image
• Due to the Platform Attitude (orientation):
Roll, Pitch, and Yaw
• Due to the Sensor’s Design Characteristics:
Tangential, V/H
roll distortion….
Digital Imaging and Remote Sensing Laboratory
Roll DistortionRoll Distortion
hypothetical output image
Nadir
direction of flight
Roll Distortion introduces a shift in the
acquired scan lines. tangent distortion….
Digital Imaging and Remote Sensing Laboratory
h y p o th e ti c al o u tp u t im a g e
direction of flight
DnadirDedge
Tangent DistortionTangent Distortion
- All pixels collected at equal angular increments.
- Unequal ground representation between the collected
ground pixels (Dedge > Dnadir) V/H distortion….
Digital Imaging and Remote Sensing Laboratory
Oversampling….
• V/H - ratio of aircraft velocity and altitude
• oversampling - scan rate is too fast when
compared to the ratio
• undersampling - scan rate is too slow when
compared to the ratio
V/H DistortionV/H Distortion
Digital Imaging and Remote Sensing Laboratory
hypothetical output image
Nadir
direction of flight
V/H Distortion - oversamplingV/H Distortion - oversampling
Undersampling….
Scan lines are acquired too fast, causing an overlap
Digital Imaging and Remote Sensing Laboratory
V/H Distortion - undersamplingV/H Distortion - undersampling
hypothetical output image
Nadir
direction of flight
proposed hypothesis….
Scan lines are acquired too slow, leaving gaps
Digital Imaging and Remote Sensing Laboratory
HypothesisHypothesis
Develop a set of C algorithms to correct the
discussed geometric distortions in line
scanner imagery.
Research Progress….
Digital Imaging and Remote Sensing Laboratory
Uncorrected InputUncorrected Input
512
512
Scan Parameters
512 scan lines
512 pixels per line
altitude = 1000 ft
Field of View = 45o
Digital Imaging and Remote Sensing Laboratory
Corrected OutputCorrected Output
334
551
• Roll Correction - (shifting of scan lines rectifies the image)
• V/H Correction - (eliminated the stretching in scene objects)
• Tangent Correction - (pixels represent equal ground areas)
Digital Imaging and Remote Sensing Laboratory
• Nearest Neighbor Resampling
• Mean Radiometry (DC Distribution) Preserved
Histogram ComparisonHistogram Comparison
Input Image (512 x 512) Output Image (551 x 334)
Digital Imaging and Remote Sensing Laboratory
Uncorrected InputUncorrected Input
512
512
Scan Parameters
512 scan lines
512 pixels per line
altitude = 250 ft
Field of View = 45o
Distortions:
V/H (Undersampling)
Tangent
Digital Imaging and Remote Sensing Laboratory
Corrected OutputCorrected Output
1336
541
• V/H Correction - (eliminated compression
effects in scene objects)
• Tangent Correction - (pixels represent
equal ground areas)
• Equal ground representation along and
across track (square ground pixels)
Digital Imaging and Remote Sensing Laboratory
Input Image (512 x 512) Output Image (541 x 1336)
Histogram ComparisonHistogram Comparison
• Nearest Neighbor Resampling
• Mean Radiometry (DC) Preserved
Research Progress….
Digital Imaging and Remote Sensing Laboratory
Uncorrected InputUncorrected Input
512
512
Scan Parameters
512 scan lines
512 pixels per line
altitude = 250 ft
Field of View = 45o
slant path (45 degrees)
Distortions:
V/H (Undersampling)
Tangent (ex:curvature
along the diagonal)
Digital Imaging and Remote Sensing Laboratory
Corrected OutputCorrected Output
615
1025
• V/H Correction - (eliminated compression
effects in scene objects)
• Tangent Correction - (pixels represent
equal ground areas)
• Improvement in the appearance of the
diagonal
Digital Imaging and Remote Sensing Laboratory
Algorithm ImprovementsAlgorithm Improvements
• Single resampling after Roll, Tangent, and V/H
corrections limits further image degradation (ex:
blurring due to bilinear resampling)
• Choice of Resampling Algorithms:
Nearest Neighbor or Bilinear
• Corrects multiple bands simultaneously
Error Analysis….
Digital Imaging and Remote Sensing Laboratory
Scientific AnalysisScientific Analysis
• accurate determination of ground pixel’s position dictates
the appearance of the resulting image.
• determines which flight parameters are the largest
sources of error : aircraft velocity, aircraft altitude, roll
angle, start angle, IFOV?
How ? ...
Digital Imaging and Remote Sensing Laboratory
Error Sensitivity AnalysisError Sensitivity Analysis
Based on the governing equations, estimate the errors across track (X) and along track (Y)
Starting ofCollection
Y (Along Track)
X (Across Track)
Error Across Track….
Digital Imaging and Remote Sensing Laboratory
Error Across TrackError Across Track
direction of flight
DnadirDedge
H
start + roll
IFOV
ground pixel ctr[x] = H * tan ( start + roll + (IFOV*x) + (IFOV/2) )
Plots….
Digital Imaging and Remote Sensing Laboratory
Roll Angle EffectsRoll Angle Effects
Across Track Error due to Roll Angle(altitude=1000 ft)
(roll=0 deg.)
0
5
10
15
20
25
30
35
40
45
500.
000
0.10
3
0.20
6
0.30
9
0.41
3
0.51
6
0.61
9
0.72
2
0.82
5
0.92
8
1.03
1
1.13
4
1.23
8
1.34
1
1.44
4
1.54
7
1.65
0
1.75
3
1.85
6
1.96
0
2.06
3
Uncertainty in Roll Angle (degrees)
Err
or
(pix
els)
edge pixel (45 deg.)
nadir pixel (0 deg.)
• Largest source of error across track
• Sensor’s ability to accurately determine a ground pixel’s
position decreases for pixels near the edges
Digital Imaging and Remote Sensing Laboratory
Altitude EffectsAltitude Effects
Across Track Error Due to Altitude(altitude=1000 ft)
(roll=0)
0
5
10
15
20
250.
00
1.81
3.63
5.44
7.25
9.07
10.8
8
12.6
9
14.5
1
16.3
2
18.1
3
19.9
5
21.7
6
23.5
7
25.3
8
27.2
0
29.0
1
30.8
2
32.6
4
Uncertainty in Altitude (feet)
Err
or
(pix
els)
edge pixel (45 deg.)
nadir pixel (0 deg.)
• Second Largest source of error across track
• Nadir pixel unaffected by the sources of error
Error Along Track….
Digital Imaging and Remote Sensing Laboratory
Error Along TrackError Along Track
Nadir
Velocity
yH
ny= n * ( (H *IFOV) - (Velocity * time_per_scan) )
Plots….
Digital Imaging and Remote Sensing Laboratory
Along Track Error Due to Uncertainties in Velocity(altitude=1000 ft)
(velocity=166 ft/sec)(scan rate=30 Hz)
0
5
10
15
20
250
1.62
3.24
4.86
6.48 8.1
9.72
11.3
4
12.9
6
14.5
8
16.2
17.8
2
19.4
4
21.0
6
22.6
8
24.3
25.9
2
27.5
4
29.1
6
30.7
8
32.4
Uncertainty in Aircraft Velocity (ft/sec)
Err
or
(pix
els)
1 pixel from nadir6 pixels from nadir32 pixels from nadir
Velocity EffectsVelocity Effects
• Largest source of error along track
• Position Error increases with aircraft velocity
Digital Imaging and Remote Sensing Laboratory
Along Track Error due to Altitude(altitude=1000 ft)
(velocity=110 mph)(scan rate=30 Hz)
0
2
4
6
8
10
12
14
16
18
0.00
1.63
3.26
4.90
6.53
8.16
9.79
11.4
2
13.0
5
14.6
9
16.3
2
17.9
5
19.5
8
21.2
1
22.8
5
24.4
8
26.1
1
27.7
4
29.3
7
31.0
1
32.6
4
Uncertainty in Altitude (ft)
Err
or
(pix
els)
near nadir (0.1 deg.)
edge pixel (45 deg.)
Altitude EffectsAltitude Effects
• Second Largest source of error along track
Future Work….
Digital Imaging and Remote Sensing Laboratory
Conclusion - Future WorkConclusion - Future Work
• Correction algorithms successfully improve the
visual appearance of the image
• Incorporate the algorithms with (MISI) line
scanner
• Incorporate other geometric distortions, such as
pitch and yaw
Acknowledgments….
Digital Imaging and Remote Sensing Laboratory
AcknowledgmentsAcknowledgments
I’d like to thank the following people for contributing to this
research:
• Dr. John Schott
• Bryce Nordgren
• Scott Brown
• Rolando Raqueño
www.cis.rit.edu/~pak2930