lecture 8 2017 - welcome | cirescires.colorado.edu/esoc/sites/default/files/class-files...density...
TRANSCRIPT
GEOG4110/5100AdvancedRemoteSensing
Lecture8
1GEOG4110/5100
• GeometricEnhancement:Richards5.1- 5.8• GeometricProperties5.10
Image-to-ImageContrastMatching
GEOG4110/5100 3
XValuesSource
CEqual.
yMod.Values
Neary
0 0 0 0
1 0 0 0
2 0 0 0
3 0 0 0
4 0 0 0
5 1 1 1
6 3 1.8 2
7 5 2.6 3
8 6 3 3
9 7 4 4
10 8 8 5
11 8 8 5
LUTforcontrastmatching
HistogramMatching
GEOG4110/5100 4
Landsat MSSscenesofnorthernsuburbsofSydneyAustralia(fromRichardsandJia,2006)
Autumn Summer
GEOG4110/5100 5
HistogramMatchingAutumnaftermatchingto
summerhistogram Summer
Landsat MSSscenesofnorthernsuburbsofSydneyAustralia(fromRichardsandJia,2006)
DensitySlicing
GEOG4110/5100 6
• Maprangesofbrightnesstoparticularshadesofgrayorcolor– Losesdetail– Reducesnoise
• Thisisasimpleone-dimensionalparallel-pipedclassifier(Ch.8)
• Sevengraylevels(below)orsixdistinctcolors(right).Couldbemoreorlessineithercase
GEOG4110/5100 7
Contouringinwaterdetailtodefinebathymetryusingdensityslicing.Upperleft:Landsat MSScompositeimageofbands5and7smoothedtoreducelinestripinglowerleft:blackandwhitedensityslicing;lowerright:colordensityslicing
GeometricEnhancement• Enhancesgeometricdetailinanimage,asopposedto
radiometricdetail.• Changesinpixelbrightnessaredrivenbygeometric
considerations,andthusaredirectlyinfluencedbythecharacterofothersurroundingpixels.– Spatialinterdependenceofpixelvaluesleadstovariationsinthe
perceivedimagegeometricdetail– Operationsoccuroverneighborhoods
• Ourcurrentfocuswillbeintheimagedomainasopposedtothespatialfrequencydomain.– Imagedomain:operationsconsiderthecharacteristicsoftheimage
itself– Spatialfrequency:operationsconsiderrateatwhichimageintensity
valuesarechangingintheimagedomain
GEOG4110/5100 8
GEOG4110/5100 9
Convolution
(ConvolutionFilter)
Convolution:Applicationofafunctiontoasignalthatmodifiesthesignal
h(t) y(t)f(t)
GEOG4110/5100 10
Convolution
(ConvolutionFilter)
Convolution:Applicationofafunctiontoasignalthatmodifiesthesignal
h(t) y(t)f(t)
https://en.wikipedia.org/wiki/Convolution
Convolution:A functionderivedfromtwogivenfunctionsbyintegrationthatexpresseshowtheshapeofoneismodifiedbytheother.
• Convolutionfilterproducesoutputimageinwhichthebrightnessvalueatagivenpixelisafunctionofthebrightnessvaluesoftheneighboringpixels.
• Convolutionfiltersinclude:LowPass,HighPass,Median,Sobel,Roberts,etc…
GEOG4110/5100 11
Convolution
ConvolutionFilter),(),( nmtnm ),( jir
Convolution:Applicationofafunctiontoasignalthatmodifiesthesignal• Imageprocessing:convolutionseekstoachieveanintendedoutcome• Sensingsystems:producesanunavoidableoutcomeforwhichwemay
seektocompensate
f
GeometricEnhancementTemplate
GEOG4110/5100 13
ForanyMx Npixelsizedtemplate,theresponseforimagepixeli,j is:
Where:
f(m,n)isthepixelbrightnessvalueaddressedaccordingtothetemplateposition,and
t(m,n)isthetemplateentryatthatlocation(canbesimilartoaweighting)
= =
=M
m
N
nnmtnmjir
1 1),(),(),( ∑∑f
ImageSmoothing(LowPassFiltering)• Reduceshigh-variabilityvalues(suchasnoise)inanimageby
“sliding”asmoothingtemplateacrosstheimage– Example:MeanValueSmoothing
– Pixelr(i,j)isassignedtheaverageofallvalueswithinthetemplate
GEOG4110/5100 14
r(i, j) =1MN
(m,n)n=1
N
m=1
M
∑∑f
MeanSmoothingFilters
GEOG4110/5100 15
OriginalImage 3x3meansmoothingfilterapplied
5x5meansmoothingfilterapplied3x1meansmoothingfilterapplied
Originaltimeseries
Smoothedtimeseriesusinginterval3 Smoothedtimeseriesusinginterval5GEOG4110/5100 16
TimeSeriesSmoothingUsingMovingAverage
0
20
40
60
80
100
120
140
160
0 20 40 60 80 100
Precip.(mm)
Day
0
20
40
60
80
100
120
140
0 20 40 60 80 100
Precip.(mm)
Day
0
20
40
60
80
100
120
140
0 20 40 60 80 100
Precip.(mm)
Day
ImageSmoothing(LowPassFiltering)
• Meansmoothingwillbluredgeswithinanimage.Toavoidthis,wecanapplyathresholdtothefilter
– TheThresholdfilterleaveslargegradientslargelyunchanged
– Usuallypre-specifiedbasedonaprioriknowledgeofimageproperties
GEOG4110/5100 17
(i, j) =1MN
(m,n)n=1
N
m=1
M
r(i,j)=r(i,j)if|f(i,j)– r(i,j)| <T
r(i,j)=f(i,j)if|f(i,j)– r(i,j)| ≥T
∑∑fr
MedianFiltering• Ratherthanassign
themeanofvalueswithinthetemplatetothecenter(i,j)pixel,themedianisassigned.– Avoidsaveragingin
sporadicnoisydata
GEOG4110/5100 18
OriginalImage
OriginalPlusnoise
MedianFilteredImage
EdgeDetectionImageEnhancement• EdgeEnhancementincreasesgeometricdetailinanimage
– Edgesareverysharpgradientsinbrightnessindicatingboundariesoffeaturesinanimage
– Accomplishedbydetectingedgesandaddingthembacktotheoriginalimagetoincreasecontrastorbyusingsaturatedoverlays(blackorwhite)ontheoriginalimagetodefineborders.
– Whydowecareaboutedges???
• Threegeneralapproachestoedgedetection– Useofanedge-detectiontemplate– Subtractingasmoothedimagefromitsoriginal– Calculatingspatialderivatives(spatialgradients)– Firsttwoaredesignedtoexaggerateinterfaces,lastisdesignedto
quantifytransitions
GEOG4110/5100 19
LinearEdgeDetectingTemplates-1 0 +1
-1 0 +1
-1 0 +1
GEOG4110/5100 20
t(m,n) =Templatethatdetectsverticaledgesinanimageisgivenby
Centralvalueistheaccumulateddifferencehorizontallybetweenpixelsin3adjacentrows
2 2 2 2 8 8 8 8
2 2 2 2 8 8 8 8
2 2 2 2 8 8 8 8
2 2 2 2 8 8 8 8
2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2
� � � � � � � �
� 0 0 18 18 0 0 �
� 0 0 18 18 0 0 �
� 0 0 12 12 0 0 �
� 0 0 6 6 0 0 �
� 0 0 0 0 0 0 �
� 0 0 0 0 0 0 �
� � � � � � � �
LinearEdgeDetectingTemplates
+1 +1 0
+1 0 -1
0 -1 -1
GEOG4110/5100 21
Vertical Horizontal Diagonal Diagonal
Centralvalueishorizontallyaccumulateddifference
-1 0 +1
-1 0 +1
-1 0 +1
-1 -1 -1
0 0 0
+1 +1 +1
0 +1 +1
-1 0 +1
-1 -1 0
NW/SEEdge NE/SWEdge
Centralvalueisverticallyaccumulateddifference
CentralvalueisaccumulateddifferenceacrossaNW/SEline
CentralvalueisaccumulateddifferenceacrossaNE/SWline
• Thetemplateisreferredtoasakernel• Thesystematicsequentialapplicationofthattemplateacrossthe
imageisreferredtoasconvolutionofthekernel• Otherconvolutionkernelsfordifferentapplicationsaregivenbelow
SpatialDerivativesTechniques:GradientOperators
• Consideracontinuousbrightnessfunctioninsteadofadiscretebrightness
• Wecandefinethespatialgradient()as:
GEOG4110/510022
Discrete Continuousx
y
Where:f(x,y) isthebrightnessvalueatpixellocationx,y andiandjareunitvectorsinthex andy directionsrespectively
Δ
Δ
(x,y) =x(x,y)i +
y(x,y) jd
df d
dff
SpatialDerivativesTechniques:GradientOperators
GEOG4110/510023
Discrete Continuousx
y
Thespatialgradientameasureofhowabruptachangeisinagivendirection.Itisdefinedasthevectorsumofchangewithrespecttothex directionandchangewithrespecttotheydirection,takeninthedirectionofmaximumgradient
Δ
(x,y) =x(x,y)i +
y(x,y) jd
df d
dff
SpatialDerivativesTechniques:GradientOperators
• Foredgedetection,wetypicallyfocusonthemagnitudeanddirectionofchangegivenrespectivelyby:
• Inotherwords,themagnitudeofthevectoristhevector(Pythagorean)sumofthegradientinthexdirectionandthegradientintheydirection
• Theaboveisforcontinuousgradients.Fordiscretegradients(i.e.acrosspixelsinimagery),wereplacethederivativeswithdifferences. Twodifference-basedspatialoperatorswewilldiscussaretheRobertsoperatorandtheSobeloperators(eachisafunctioninENVI)
GEOG4110/5100 24
1 =x(x,y)
2 =y(x,y)
∇ = ∇1 +∇2
Where
2 2 =tan-1(/)Δ Δ Δ
2 1and
∇ ∇ddf d
df
TheRobertsOperatorDiscretecomponentsofthederivativeonthepreviouschartaregivenby
forthepointi+½,j+½
Inotherwords:weassessthegradientacrossthetwodiagonalsasameanstodeterminetheedges
GEOG4110/5100 25
1 = (i, j) (i +1, j +1)
2 = (i +1, j) (i, j +1)
i+1
j j+1
(i+½,j+½)
Sincealocalgradientiscomputed,itisnecessarytospecifyathresholdvaluetodetermineedgegradientsandsuppressminorgradients
i
Detectshorizontal,verticalanddiagonaledgesandassignsthemtotheedgethatisinthedirectionofincreasingIandj,offsetbyhalfapixel.
∇ f - f ∇ f - f