On-Orbit MTF Measurement and Product Quality Monitoring

On-Orbit MTF Measurement and Product Quality Monitoring for Commercial Remote Sensing Systems March 16th, 2006Steven Person

https://ntrs.nasa.gov/search.jsp?R=20070038210 2019-08-29T18:38:25+00:00Z

brought to you by C

OR

EV

iew m

etadata, citation and similar papers at core.ac.uk

provided by NA

SA

Technical R

eports Server

https://core.ac.uk/display/42754722?utm_source=pdf&utm_medium=banner&utm_campaign=pdf-decoration-v1

On-Orbit MTF Measurement and Product Quality Monitoring March 16, 2006 2

Outline

Problem Definition

Technique Overview

Validation

Current Software Implementation

Product Quality Monitoring Architecture


Problem Definition

Determine the MTF of an on-orbit satellite using in-scene targets:– Slant Edge

– Line Source

– Point Source

– Radial Target

Attempt to facilitate the MTF calculation by automatically locating targets of opportunity.

Incorporate MTF results into a product quality monitoring architecture.


Relation Between MTF Components

x∂∂ ℑ

∫ ∞−

x

OPTICAL SYSTEM

Input ESF LSF MTF1−ℑ

Initialization and opportunistic targets are chosen that represent the MTF in the spatial domain.

Ideal targets have simple mathematical relationships.


Review of Potential Targets for MTF Calculation

Method Approach Advantage DisadvantageEdge

Gradient

Computes LSF from edge profile

Basic approaches are similar but different in ways edge profile is determined

ISO has a standard approach

Less sensitive to alignment issues

Targets easier to implement

Good energy at all frequencies

Typically uses curve fits for edge profiles

Computes LSF indirectly from ESF and uses differentiation

Can introduce noise

Pulse

Input

Computes LSF directly from target Less numerical error from MTF Requires knowledge of target width and resolution for reliable results

Point

Source

Computes point spread in x & y directions as a function of intensity and distance across imaged point

Provides 2-D MTF Requires confidence about location of point source center

Multiple aligned points necessary

Various signal-to-noise issues (atmos. effects, neighboring points, single point SNR, etc.)

Radial

Target

Analyzes a series of “pulses” lying on concentric paths about a circle

Can provide visual quality assessment

Provides contiguous frequencies

Difficult to implement

High potential for aliasing


Two Stage Algorithm

Input image area is sequentially searched for areas of edge content.

A set of user modified parameters are defined to constrain located edges:

– Edge Size

– Edge Angle

– Contrast

– Uniformity of light and dark areas

Edges that satisfy all criteria are projected to 1D and passed to the MTF algorithm.

MTF is calculated using a method developed by B. Tatian JOSA, Vol. 55, pp. 1014-1019.

Avoids taking a discrete derivative of the ESF by approximating the MTF as a set of trigonometric series.

Errors in the algorithm are dominated by:

– Edge Size

– Angle

– SNR

Edge Finding MTF Calculation


MTF Algorithm Flow

Area Selected for Processing

Estimate Edge Location

Calculate Edge Angle

Pass?

Project to 1D ESFNormalize ESF Pass?

Calculate Edge MTF

Add to Composite Database

Proceed to Next Area

Results are averaged in the frequency domain.

Outliers of the distribution rejected from the final average.


Validation of Current method

Four images of the Big Spring, TX test target were provided to ITT by DigitalGlobe.

MTF algorithm was verified by manually selecting one along-scan and one cross-scan edge from each image for processing.

Full algorithm was used to process a 400x400 pixel area.

Imagery courtesy of DigitalGlobe/Reprinted with Permission


Results of Manual MTF Estimate from Edge Target

Along-Scan Manual Estimates

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50

Freq. (cycles/pixel)

MTF

26-Dec-0408-Jan-0513-Jan-0521-Jan-05

Cross-Scan Manual Estimates

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF

26-Dec-0408-Jan-0513-Jan-0521-Jan-05

Average Along-Scan MTF

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF

AveragePlus 95% Conf.Minus 95% Conf.

Average Cross-Scan MTF

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF



Example Run on Test Target Crop



Automatic MTF Estimate from Edge Target

Along-Scan Auto Estimates

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF

26-Dec-0408-Jan-0513-Jan-0521-Jan-05

Cross-Scan Auto Estimates

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF

26-Dec-0408-Jan-0513-Jan-0521-Jan-05

Average Cross-Scan MTF

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF


Cross-Scan Average MTF

00.10.20.30.40.50.60.70.80.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF



Comparison to DigitalGlobe Results

Good agreement between the manual estimation and automatic estimation when compared to independent DG results.

Positive bias in the low frequencies due to small edge size used in computation.

DG vs. ITT MTF ComparisonAlong-Scan

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF

DG MTFManual MTFAuto MTF

DG vs. ITT MTF ComparisonCross-Scan

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0.00 0.10 0.20 0.30 0.40 0.50


MTF

DG MTFManual MTFAuto MTF


Example Operational Image

Algorithm executed with a nominal parameter set on image with potential edge content.

Red squares indicate targets used to estimate along-scan MTF.

Green squares indicate targets used to estimate cross-scan MTF.



Results from Extracted Test Image

Cross-scan bias observed in individual edges used to estimate MTF.

– Possibly due to unobserved roof structure.


Roof Edge ExampleManual Edge Crop


Further investigation shows nothing unusual about edge.

Presents a difficult problem for automatic edge detection routine.


Prototype Software Implementation

The current software implementation attempts to address the edge selection issues.

Coded in IDL as a plug-in to ENVI.

Allows user to manipulate imagery with built-in ENVI functionality and select Regions of Interest within an image where edge content appears.

ROIs are imported into a separate GUI for processing and result display.


Screen Shot of MTF Measurement Toolkit

Facilitates quick processing of data for time critical results or repetitive monitoring.

Using a system with a known MTF the Toolkit can identify regions around the globe that approximate “ideal edges”.



Areas for Future Improvement and Investigation

Investigate alternate edge finding techniques.– Methods which include the least false positives.

– Region growing techniques to increase number of samples along an edge.

Investigate benefit of aggregation in the spatial domain vs. the frequency domain.

Incorporate a database function that allows for tracking and trending of results.


Relative Edge Response Relation to Image Quality

0 1 3 42-4-3 -1-2Pixel location

Nor

mal

ized

Edg

e R

espo

nse

1

0.2.4.6.8

The slope of this line is the RER.

( ) ( ) ( )SNRGAHARERbGSDaANIIRS 210 loglog −−+−= ( )RERb log+NIIRS=

RER is easily calculated along with the MTF using the same algorithm.

RER is the second largest contributing factor to the General Image Quality Equation (GIQE).


Two possible PQ Monitoring Architectures

Develop full GIQE model to monitor NIIRS ratings.

Flag images which fall Nσoutside the historical distribution of NIIRS ratings.

Identify shifts in histogram mean.

Monitor each parameter of the GIQE separately.

Calculate a baseline mean with confidence bounds.

Indicate when a parameter falls outside of the baseline behavior.

Historical RER Trend Analysis

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1 3 5 7 9 11 13 15 17 19

Image Sample

Rela

tive

Edge

Res

pons

e

Distribution of NIIRS Rating

0

200

400

600

800

1000

2 3 4 5 6 7 8

NIIRS Rating

Imag

e Co

unt

Documents

On-Orbit MTF Measurement and Product Quality Monitoring