MULTITEMPORALIMAGEANALYSIS:

PRACTICALLECTURE
LorenzoBruzzone
Email:lorenzo.bruzzone@ing.unitn.it
Webpage:http://rslab.disi.unitn.it

LorenzoBruzzone

Outline
1

ProcessingChainforChangeDetection

ChangeVectorAnalysis(CVA)
EffectsofRadiometricDifferences
EffectofResidualMisregistration
DetectionofMultipleChanges

CompressedChangeVectorAnalysis(C2VA)

EffectsofRadiometricDifferences
EffectofResidualMisregistration
DetectionofMultipleChanges
AnalysisofDirectionAmbiguity

LorenzoBruzzone

ProcessingChainforChangeDetection

X1
t1 image

Radiometric
Corrections

Image
Coregistration

ImageComparison
(CVAorC2VA)

ChangeInformation
Extraction
ChangeDetection
Map

XN
t2 image

LorenzoBruzzone

ChangeVectorAnalysis(CVA)
Assumption:only2spectralchannelsareconsideredforeachdate.
Magnitude
Image

Pixel(i,j)
XX11
X2
Multispectral
imaget1

XX21
Multispectral
imaget2

Vector
Difference

XXD1
Multispectral
differenceimage

X2
X1

XD

X X
2

2, D

1, D

X
X tan 1 1,D
X 2,D

Direction
Image

X1

LorenzoBruzzone

CVA
Selectimageattimet1;
Selectimageattimet2;

Selectthedesiredpreprocessingsteps;

SelectCVAoption;
Selectthespectralchannelstohighlightthe
changes;
PressCalculateto:
Performpreprocessing(ifany);
Computemagnitudeanddirection;
Visualizepolarrepresentation.
LorenzoBruzzone

CVA
IfCoregistrationcheckbox
isselected,coregistration
formappears;
SelectGroundControlpoints
(atleast4);
Coregistrationisperformed
bymeansofbilinear
interpolation.
Oncefinishedselect
File -> Close Control
Point Selection Tool

LorenzoBruzzone

CVA

Polarplotexample
LorenzoBruzzone

CVA
Select threshold for the magnitude variable:
Perform automatic threshold selection
(press Compute automatic threshold and
wait some seconds);
Manual threshold selection moving the red
line on the histogram or cursor on the T
scroll;
Use zoom to better detect the threshold.
PressApplythresholdto:
Thresholdthemagnitude;
Movetodirectionanalysis.
Select whether presence/absence of change information
should be used in the analysis of direction information.
LorenzoBruzzone

CVA
Selectthreshold(s)forthedirectionvariable:
Manualthresholdselectionmovingthered
andblacklinesonthehistogram.
Ormovingthecursors ontheT1 andT2 scrolls;
Use zoom option to better detect the
threshold(s).

PressApplythreshold(s)to:
Thresholdthedirection;
Visualizethefinalchangedetectionmap.

LorenzoBruzzone

CVA

Changedetectionmapexample
LorenzoBruzzone

CVA:Outputs&Notes
Outputs:

Multispectraldifferenceimage;
Magnitudeimage;
Directionimage;
Thresholded magnitudeimage;
Thresholded directionimage;
Changedetectionmap;
Polarplot;

Notes:
AlloutputsaresavedinOutputfolder.
ImagesaresavedasadatafileplusheaderformatandcanbeloadedeitherinMatlab
orENVIenvironment.
Eachrunoverwritesoutputfiles,ifyoudontwanttoloosethemrenamethefiles.
PolarplotissavedinMatlab .figformat.
LorenzoBruzzone

CVA:SuggestedTests
Preformtrialsinordertoanalyze:
Theeffectofusingimageswithradiometricdifferences;
Theeffectofusingimageswithasignificantamountofresidualmisregistration;
Theeffectofusingdifferentcombinationsofspectralchannels.
Theeffectofthepresenceofmultiplechanges.

Tip: insteadofthresholding,manuallydrawregionsonthepolarplottoisolatespecific
clustersofpixels.

LorenzoBruzzone

CVA:ExampleandDiscussionofResults
Studyarea:LakeMulargia,SardiniaIsland(Italy).
Multitemporaldataset:aportionof412300pixelsoftwoimagesacquiredbytheTM
sensorofLandsat5satelliteinSeptember1995andJuly1996.

BeforeChange

AfterChange

ReferenceMap

LorenzoBruzzone

CVA:ExampleandDiscussionofResults
CorrectedImages(Idealcase)

Optimalthresholdvalueonthe
magnitudevariable:idealcase

Registrationnoiseeffects

Registrationnoiseeffects

Radiometricdifferenceeffects

Thresholdvalueonthemagnitude
variable:radiometricdistortioncase

F.Bovolo,L.Bruzzone,ATheoreticalFrameworkforUnsupervisedChangeDetectionBasedonChangeVectorAnalysisinPolarDomain,
IEEETransactionsonGeoscienceandRemoteSensing,Vol.45,No.1,2007,pp.218236.

LorenzoBruzzone

CompressedChangeVectorAnalysis(C2VA)
CVA in 2 dimensions permits to easily visualize the change information in polar
coordinates, but may results in the loss of information due to spectral channel
selection.
CVA may be applied on B > 2 spectral channels in hyperspherical coordinates.
However, when B is greater than 3 it is impossible to visualize the data in the polar
domain.
Compressed CVA (C2VA) can overcome the abovementioned limit of polar CVA.

15

LorenzoBruzzone

CompressedChangeVectorAnalysis(C2VA)
X11
X

Magnitude:thelengthofthemultispectral
differencevector(XD).

BD Multispectral
image t1

XXD1

+
BD Multispectral
difference image
={n, c}
c ={ c1 ,c2 ,...,cK }

X21
X
BD Multispectral
image t2

2
b,D

b 1

b ,2

Xb ,1

b 1

bth spectralbandofX1

Direction:theanglebetweenthemultispectraldifferencevector(XD)andareferencevector
(Xref)inaBD space.
X2,D

B
X arccos Xb,D Xb ,ref
b1

XD

[0, ]

X3,D

Xref

X1,D

X X
2
b,D

b 1

2
b ,ref

b 1

BD unitvector

X ref B ,..., B
B
B
LorenzoBruzzone

16

CompressedChangeVectorAnalysis(C2VA)
Definitions
1. CompressedCVA(C2VA)Domain

C 2VA , : 0 max and 0

max

max

2
X b,D

b 1

k 1
k 2

max

Sk

> RandomvariableassociatetodirectionimageX

SAc

2. SemiCircleofunchangedpixels

SCn , : 0 T and 0

SCn

3. SemiAnnulusofchangedpixels

SAc , : T max and 0

4. Annularsectorofthekthkindofchange

S k , : T and k1 k2 , 0 k1 k2
LorenzoBruzzone

17

C2VA
Selectimageattimet1;
Selectimageattimet2;

Selectthedesiredpreprocessingsteps;

SelectC2VAoption;

PressCalculate
andfollowthesimilarstepsdescribed
forCVAoption.
LorenzoBruzzone

C2VA:SuggestedTests
Preformtrialsinordertoanalyze:
Theeffectofusingimageswithradiometricdifferences;
Theeffectofusingimageswithasignificantamountofresidualmisregistration;
Theeffectofthepresenceofmultiplechanges;
Theeffectofambiguityinthedirectioninformation.

Tip: insteadofthresholding,manuallydrawregionsonthepolarplottoisolatespecific
clustersofpixels.

LorenzoBruzzone

C2VA:ExampleandDiscussionofResults
Studyarea:LakeMulargia,SardiniaIsland(Italy).
Multitemporaldataset:aportionof412300pixelsoftwoimagesacquiredbytheTM
sensorofLandsat5satelliteinSeptember1995andJuly1996.
Changes:2naturalchanges,1simulatedchange
Lake surface enlargement

Simulated burned area

Open quarry enlargement

September 1995

July 1996
LorenzoBruzzone

20

10

C2VA:ExampleandDiscussionofResults
C2VA
90

CVA (bands 4 and 7)

200

90

120

60

30

150

180

200

120

60

150

30

180

Open quarry enlargement

Simulated burned area
Lake surface enlargement

180
210

330

5x10-4

240

300
270

LorenzoBruzzone

21

C2VA:ExampleandDiscussionofResults
Advantages:
C2VA permits to easily visualize the change information in polar coordinates without
the need of preselecting pairs of spectral channels (which is difficult when no prior
information on the kinds of changes present in the images is available).
Disadvantages:
Some ambiguity may rise from the dimension reduction process mainly for the
simplified representation of the angle variable: this may result in similar direction
values for different kind of changes.

LorenzoBruzzone

22

11

References
1.

W.A. Malila, Change vector analysis: an approach for detecting forest changes with Landsat,
Proc. LARS Machine Processing of Remotely Sensed Data Symposium, W. Lafayette, IN: Laboratory
for the Application of Remote Sensing, pp. 326336, 1980.

2.

A. Singh, Digital change detection techniques using remotelysensed data, International Journal
of Remote Sensing, Vol. 10, No. 6, pp.9891003, 1989.

3.

L. Bruzzone, D. Fernandez Prieto, A minimumcost thresholding technique for unsupervised

change detection, International Journal of Remote Sensing, Vol. 21, No. 18, 2000, 35393544.

4.

L. Bruzzone, D. Fernandez Prieto, Automatic analysis of the difference image for unsupervised
change detection, IEEE Transactions on Geoscience and Remote Sensing, Vol. 38, No.3, 2000, pp.
11711182.

5.

L. Bruzzone, D. Fernandez Prieto, An adaptive semiparametric and contextbased approach to

unsupervised change detection in multitemporal remote sensing images, IEEE Transactions on
Image Processing, Vol. 1, No. 4, 2002, 452466.

6.

P.R. Coppin, I. Jonckheere and K. Nachaerts, Digital change detection in ecosystem monitoring: A
review, International Journal of Remote Sensing, Vol. 25, No. 9, pp. 15651596, 2004.
LorenzoBruzzone

References
7.

D. Lu, P. Mausel, E. Brondzio and E. Moran, Change detection techniques, International Journal of
Remote Sensing, Vol. 25, No. 12, pp. 23652407, 2004.

8.

A. J. Radke, S. Andra, O. AlKofahi, and B. Roysam, Image Change Detection Algorithms: A

Systematic Survey, IEEE Transactions on Image Processing, Vol. 14, No. 3, pp. 294307, 2005.

9.

F. Bovolo, L. Bruzzone, A Theoretical Framework for Unsupervised Change Detection Based on

Change Vector Analysis in Polar Domain, IEEE Transactions on Geoscience and Remote Sensing,
Vol. 45, No. 1, pp.218236, 2007.

10. F. Bovolo, L. Bruzzone, S. Marchesi, Analysis and Adaptive Estimation of the Registration Noise
Distribution in Multitemporal VHR Images, IEEE Transactions on Geoscience and Remote Sensing,
Vol. 47, No. 8, pp. 26582671, 2009.
11. S. Marchesi, F. Bovolo, L. Bruzzone, A ContextSensitive Technique Robust to Registration Noise
for Change Detection in VHR Multispectral Images, IEEE Transactions on Image Processing, Vol.
19, No. 7, pp. 1877 1889, 2010.
12. F. Bovolo, S. Marchesi, L. Bruzzone, A Framework for Automatic and Unsupervised Detection of
Multiple Changes in Multitemporal Images, IEEE Transactions on Geoscience and Remote
Sensing, Vol. 50, No. 6, pp. 21962212, 2012.
LorenzoBruzzone

12

Acknowledgments
The material for this practical lecture, including slides and software tool, has been
developed with the collaboration of Remote Sensing Laboratory members at the University
of Trento.
Special thanks to: Carlo Marin, Francesca Bovolo, Sicong Liu.

LorenzoBruzzone

13