RS Image Corrections and Registrationraul/RS/ELE2769_03_RSImage... · G R Y RE NIR1 NIR2 Digital Sensors. Radiometric Correction •Digital numbers (pixel intensity values) are unique

RS Image Corrections and Registration Radiometric and Geometric Corrections

Raul Q. FeitosaGilson. A. O. P. Costa

Patrick N. Happ

Contents

Radiometric Calibration

Radiometric Correction

Conversion to TOA Reflectance

Atmospheric Correction

Geometric Correction

Orthorectification

Georeferencing

Cartographic Concepts

2

Remote Sensing Image Corrections

• Images recorded by sensors on satellites and aircrafts can contain errors in geometry and in the recorded spectral radiance values.

• A number of techniques have been proposed to eliminate or attenuate distracting elements of images: information noise.

• Three mains sources of errors/noise:

• Instrument errors

• Atmospheric effects

• Geometric distortions

• Before starting on the image correction techniques, we will learn how to transform intensity image/pixel values into physical quantities.

Radiometric Calibration/Correction

CCD Sensor

WorldView 2

• Systems deliver digital numbers (DN)

• Images: picture element (pixel)

• Signal (radiance) integrated

• Signal (radiance) discretized CPan

BG

RY

RENIR1

NIR2

Digital Sensors


• Digital numbers (pixel intensity values) are unique to each sensor and should not be directly compared to imagery from other sensors in a radiometric/spectral sense.

• DNs are not equivalent to spectral radiance values (W m-2 sr-1 μm-1), although the two sets of values correlate.

• To be able to compare RS data from different sensors (or even from the same sensor at different epochs) DNs must be converted to TOA (top-of-atmosphere) spectral radiance.

• TOA radiance: as the surface of the Earth is seen from space.


• Sensor detectors are characterized by different gain and offset values (radiation transfer characteristics)

• To convert from pixel values to TOA/apparent radiance for most sensors:

Ln = On + Gn PV

Ln = radiance recorded for band n

On = offset for band n

Gn = gain for band n

PV = pixel value (DN)

Landsat 5 TM O G

1 -0.1009 0.06362 -0.1919 0.12623 -0.1682 0.09704 -0.1819 0.09145 -0.0398 0.01267 -0.0203 0.0067


• Some sensors have different gain settings (e.g. Landsat, ASTER)

• This is to register as much contrast as possible and avoid saturation:

• High gain for regions with low reflectance

• Low gain for regions with high reflectance

ASTER - Maximum Radiance (W m-2 sr-1 μm-1)Band No High gain Normal gain Low Gain 1 Low Gain 2

1 170.8 427 569

N/A2 179.0 358 477

3N 106.8 218 2903B 106.8 218 2904 27.5 55.0 73.3 73.35 8.8 17.6 23.4 103.56 7.9 15.8 21.0 98.77 7.55 15.1 20.1 83.88 5.27 10.55 14.06 62.09 4.02 8.04 10.72 67.0


• Gain and offset values (and eventually other sensor specific parameters) can change over time.

• Such parameters may change over time: calibration campaigns.


• For DigitalGlobe products:

• TDI (time-delay integration) specific absCalFactor and effectiveBandwidthare delivered with the imagery in the metadata file.


• Gain settings for each specific image/band is usually recorded at the metadata

• SPOT Header file:

Scene ID: S1H1870112102714Scene centre latitude: N0434026Scene centre longitude: E0043615Spectral Mode (XS or PAN): XSPreprocessing level identification: 1BRadiometric calibration designator: 1Deconvolution designator: 1Resampling designator: CCPixel size along line: 20Pixel size along column: 20Image size in Map Projection along Y axis: 059792Image size in Map projection along X axis: 075055Sun calibration operation date: 19861115This is a multispectral imageAbsolute calibration gains: 00.86262 00.79872 00.89310Absolute calibration offsets: 00.00000 00.00000 00.00000

Radiometric CorrectionIllumination and view angle effect

• In the absence of the atmosphere, the magnitude of the radiance reflected from or emitted by a target will vary with:

• Solar elevation (illumination) angle

• Sensor elevation (view/nadir) angle

Radiometric CorrectionIllumination and view angle effect

• Assuming the ground element is Lambertian:

• α, θ available in image metadata

cos(α)

cos(θ)L’ = L ––––––

L’ = corrected radiance

L = radiance measured by sensor

α = viewing zenith angle

θ = solar zenith angle

Radiometric CorrectionConversion to TOA/apparent Reflectance

• Given the value of the spectral radiance recorded by a sensor, it is possible to calculate the TOA/apparent reflectance ρ (for each band):

π L d2

Es cos(θs)ρ = ––––––––

L = radiance

d = Earth-Sun distance (above)

Es= exo-athmospheric solar irradiance (table)

θs = solar zenith angle (image meta-data)

JD = Julian date

d = 1 - 0.01674 cos[0.9856(JD - 4)

Conversion to TOA/apparent ReflectanceLandsat ETM+ Band Bandwidth (μm) Exo-atmospheric

spectral irradiance1 0.450 -0.515 196.9 2 0.525 - 0.605 184.0 3 0.630 - 0.690 155.1 4 0.775 - 0.900 104.4 5 1.550 - 1750 22.577 2.090 - 2.350 8.207 8 0.520 - 0.900 136.8

SPOT HRV Band Centre wavelength (μm)

Exo-amosphericspectral irradiance

1 0.544 187.482 0.638 164.893 0.816 110.14

ASTER Band Bandwidth (μm) Exo-atmospheric spectral irradiance

1 0.520 - 0.600 1846.92 0.630 - 0.690 1546.03 0.780 - 0.860 1117.64 1.600 - 1.700 232.55 2.145 - 2.185 80.326 2.185 - 2.225 74.927 2.235 - 2.285 69.208 2.295 - 2.365 59.829 2.360 - 2.430 57.32

π L d2

Es cos(θs)ρ = ––––––––

L = radiance

d = Earth-Sun distance

Es= exo-athmospheric solar

irradiance

θs = solar zenith angle

JD = Julian date

d = 1 - 0.01674 cos[0.9856(JD - 4)

Conversion to TOA/apparent ReflectanceExoathmospheric solar irradiance

Atmospheric Correction

• The value recorded at any pixel location is not only a function of the ground leaving radiance of the ground element

• Atmospheric effects: absorption and scattering of radiation

ρET

πLtot = –––––– + Lp

Ltot = radiance measured by sensor

Lp = path radiance

ρ = reflectance of object

E = irradiance on object

T = transmission of atmosphere

Atmospheric Correction• Important preprocessing techniques in three cases:

• Compute the ratio between two bands (scattering x wavelength)

• Part of the signal of importance is smaller than the atmospheric component (radiance upwelling from ocean is often less than path radiance)

• Measurements made at one time (time 1) are to be compared with measurements from another time (time 2)

Atmospheric CorrectionImage based methods

• Estimation of the path radiance Lp and subtraction from the signal received by the sensor

• Two common techniques:

• Histogram minimum method

• Regression method

Atmospheric CorrectionHistogram minimum method

• Calculate histograms for all bands

• Image generally contains areas of low reflectance (water, deep shadows, dark rocks)

• Pixels covering those areas will have values close to zero in NIR bands

• Corresponding pixels in other bands will have higher values

• The smaller the wavelength, the larger the offset

• Solution: determine the offset for each band and shift its histogram

Atmospheric CorrectionRegression method

• Plot pixel values in a NIR band (band B) against the values in another band (band A)

• Compute a best-fit straight line using a regression method (least-squares)

• The offset a on the x-axis is an estimate of path radiance Lp for the particular band

• Shift histogram of band by a

• Repeat for each band (visible spectrum)

Atmospheric CorrectionRadiative Transfer Models

• There are much more complex techniques for atmospheric correction

• Take into consideration atmospheric/sensor/illumination conditions at the moment the data was recorded:

• Temperature

• Relative humidity

• Atmospheric pressure

• Water vapor/Visibility

• View/nadir angle

• Solar elevation angle


• Transformation of a remotely-sensed image so that it has the scale and projection properties of maps

• Geometric correction is required in images or products from analysis that are used to:

• transform an image to match a map projection

• locate points of interest on map or image

• bring adjacent images into registration

• overlay temporal sequences of images of the same area, perhaps acquired by different sensors.

• overlay images and maps within Geographic Information Systems (GIS)


• Transformation of a remotely-sensed image so that it has the scale and projection properties of maps

• Most corrections are performed by suppliers of RS images

• Necessary if working with raw data

Geometric DistortionsEarth Rotation Skew

• During frame acquisition time Earth rotates from west to east

• To put pixels in their correct positions relative to the ground it is necessary to offset bottom of image to the west by amount of movement during image acquisition

Geometric DistortionsPanoramic distortion

• Pixel size on ground: varies with nadir/incidence angle

Nadir angle = τ

Pixel size on ground in view direction: pv = p/cos² τ

Pixel size on ground in orbit direction: po = p/cos τ

τ = 30°, p=1m in nadir: pv= 1.33m po=1.15m

Geometric DistortionsAspect ratio

• Some sensors produce images with pixels that are not square

• Reason: IFOV larger than GSD in one direction

• Example: Landsat MSS

• IFOV = 79m

• GSD (in scan direction) = 59m

• GSD (in orbit direction) = 79m

Geometric DistortionsScan Time Skew

• Satellite orbits are inclined in the north-south direction

• Inclination increases with latitude

• Skew angle θ at latitude L is given (in degrees) by:

sin(θE)

cos(L)θ = 90o - arccos ––––––

θE = satellite heading (inclination - 90o)

Geometric CorrectionSolving Geometric Problems

• Image transformations: affine, panoramic, polynomial

• Resampling and interpolation

Geometric CorrectionResampling and interpolation


• New pixels are created

• Pixel values must be defined for new pixels

• Interpolation: approximation of a pixel's color and intensity based on the values at surrounding pixels

x 2

New pixel values definedUse existing pixel values

or


Nearest neighborhood

• Nearest neighborhood

• Bilinear (average of 2x2 neighbors)

• Bicubic (average of 4x4 neighbors)

x 2

Bilinear interpolation Bicubic interpolation


• If particular distortions can be represented explicitly, mapping functions can be specified explicitly

• Transformation matrixes

Geometric CorrectionTransformation Matrixes

• Aspect ratio and panoramic correction: scaling

r

c

s0

0s

n

e

r

c


• Earth Rotation Skew: shearing

r

c

10

1

n

e k


• Scan Time Skew: rotation (counter clockwise)

r

c

)cos()sin(

)sin(-)cos(

n

e


• When cannot or do not know how to model mathematically

• Mapping polynomials


• When cannot or do not know how to model mathematically

• Mapping polynomials:

• Need to know at least six solutions toestimate coefficients (second order)

• Solutions: ground control points (GCPs)

e = a0 + a1c + a2r + a3cr + a4c2 + a5r2

n = b0 + b1c + b2r + b3cr + b4c2 + b5r2

Geometric CorrectionImage to Image Registration

• Fitting of the coordinate system of an image to that of a second image

reference image (orthophoto) unregistered image

Geometric CorrectionImage to Image Registration

• Fitting of the coordinate system of an image to that of a second image

reference image (orthophoto) coregistered image

Geometric CorrectionExercise: Matlab practice

Coregister two images

1) Read images

reference: westconcordorthophoto.png

input: westconcordaerial.png

2) Select control points

cpselect(input(:,:,1),reference)

3) When finished: File/Save Points to Workspace

4) Create transformation function

tform = cp2tform(input_points,base_points,'polynomial',2)

5) Apply transformation

output = imtransform(input,tform)

Geometric CorrectionGeoreferencing

• Rectify images in order for it to have properties of a map

• Pixels are associated to geographic coordinates (latitude and longitude)

• Distances can be measured

• Various accuracy levels

• Various techniques:

• Registration (to other images, to control points)

• Orthorectification

• Subject of a discipline called Photogrammetry

Geometric CorrectionOrthoimage

Perspective images:relative position errors

Map/Orthoimages:parallel projection

Geometric CorrectionOrthoimage

• Geometry of map, information contents of image

• Original projection: shift of position depending upon height

• Map projection: parallel projection

• Orthorectification: projection of image pixel by pixel in the height level of the given height model to the map geometry

• Required: image orientation, digital elevation model (DEM)

Geometric CorrectionHeight/Elevation Models

• DEM (Digital Elevation Model) – height of terrain surface

• DSM (Digital Surface Model) – height of visible surface

• DTM (Digital Terrain Model) – no clear definition: sometimes used as DEM, sometimes as DSM

Ortoimage

Digital Surface Model (DSM)

Digital Elevation Model (DEM)

Heights (DSM-DEM)

Geometric CorrectionHeight/Elevation Models

SRTM (Shuttle Radar Topography Mission)

28/10/2001

Mapped 80% of Earth’s

surface (February 2000)

Spatial resolution of

90m (30m in USA)

ASTER GDEM (2009)

30m resolution

Geometric CorrectionOrthorectification

projection center

ground element

image

topographicsurface (DEM)

referenceplane (map)

position of ground elementon image (projected)

correct positionof ground element

dl – error in horizontal position

dl

dh

dh – height of ground element

dl = dh x tan ν

Orthorectification

• Projection of image pixel by pixel to the map geometry

• Algorithm:

1. Choose position (xo, yo) in orthoimage

2. Compute zo (height) based on DEM

3. Calculate pixel position (xi, yi) in image corresponding to (xo, yo, zo)

4. Place intensity value from (xi, yi) to (xo, yo)

orthoimage

raw image

DEM

Orthorectification

Remaining problems:

• Objects with vertical lines (bridges, buildings)

• Geometry only correct for objects located in the height level of the DEM (for example roof tops shifted)

Orthorectification

Error in position due to DEM resolution:

Real terrain surface

DEM

dl = dh x tan ν

dl

dh

Orthorectification

Error in position of objects above DEM:

Orthorectification

High buildings, roof tops are shifted:

Orthorectification

Bridges over valleys:

• River in correct position

• Bridge shifted

Orthorectification

Bridges over Bosporus:

Orthorectification

Real orthoimage requires height information of all objects (DSM):

Geometric CorrectionCorrection Levels

projection center

h

raw image (only radiometric correction + inner sensor geometry)

image projected over DEM (orthophoto)

image projected over plane with constant height

image projected over reference surface (rough DEM)

(QuickBird Basic)

IKONOS GeoQuickBird Standard


Providers of RS images sell products of different levels of correction:

IKONOS


QuickBird

Georreferencing

World files:

• Plain text data file used by geographic information systems to georeference images

World FilesA: x component of the pixel width (x-scale)D: y component of the pixel width (y-skew)B: x component of the pixel height (x-skew)E: y component of the pixel height (y-scale)

Georeferencing

World files:

• Plain text data file used by geographic information systems to georreference images

• Naming convention (file extension)

• append the letter "w" to the end of the raster filename: image.jpg → image.jpgw

• second letter of the original filename extension is removed, and the letter "w" is added at the end:

image.jpg →image.jgwimage.tif → image.tfw

• Does not specify a coordinate system (projection; ellipsoid; datum)

Georeferencing

GeoTIFF:

• TIFF (Tagged Image File Format) image file

• Standardized tags at the header of TIFF file

Georeferencing

GeoTIFF (example):

• Projection: UTM zone 60

• Datum: WGS84

• Upper-left corner coordinates (UTM): 350807.4m, 5316081.3m

• Pixel scale/size: 100 meters

Georeferencing

GeoTIFF (example):• Projection: Lambert Conformal Conic• Datum: NAD27 • Central meridian: 120 degrees West• Pixel scale/size: 1000 meters (at origin)• Standard parallels: 41”20’ N and 48”40’ N• Latitude of the origin: 45 degrees North

Cartographic ConceptsCartographic Projection

• Method by which each point on Earth is mapped to a point on a map

• There are a variety of ways of projecting on a plane the geographical objects that characterize the Earth's surface


AzimuthalProjections

(onto a plane)

CylindricProjections

ConicProjections


• Method by which each point on Earth is mapped to a point on a map

• There are a variety of ways of projecting on a plane the geographical objects that characterize the Earth's surface


Conformal or orthomorphic projections:

• Preserve angles

• Examples: UTM, Lambert Conformal Conic

Equivalent or isometric projections:

• Preserve areas

• Example: Albers Conic

Equidistant projections:

• represent distances in true magnitude along certain directions

• Examples: Cylindrical Equidistant







Policonic


UTM (Universal Transverse Mercator)

Cartographic ConceptsUTM Projection

• Most used in high resolution orthoimages (fine scale maps)

• Coordinates expressed in meters (e.g., 655103,74; 7461340,00)

Cartographic ConceptsUTM Projection


Cartographic ConceptsEllipsoid

Earth’s surface approximation: ellipsoid of revolution (spheroid)

Cartographic ConceptsEllipsoid

Earth’s surface approximation: ellipsoid of revolution (spheroid)

Cartographic ConceptsGeodetic Datum

Cartographic data are associated to a Geodetic Reference System: Ellipsoid + Datum (point of origin)

Datum #2

Earth’s Surface

Ellipsoids

Datum #1


• Horizontal Datums most used in Brazil:

• SAD69 (Chuá, MG): used until the 2000’s

• Córrego Alegre (MG): first oficial datum

• WGS84: datum used by GPS and most RS images providers

• SIRGAS: Sistema de Referência Geocêntrico para as Américas, new official datum.

• Vertical Datum:

• Marégrafo de Imbituba (SC)


Referencing to different datums: positioning errors

Cartographic ConceptsGeodetic Reference Systems

• There are mathematical methods for re-projecting and transforming between Geodetic Reference Systems (GRS)

• Important is to know in which GRS and projection the data was provided!

Chapter 4 “Preprocessing of Remotely-Sensed Data” of the book “Computer Processing of Remotely-Sensed Images”, Mather & Koch, 4th Edition.

Richter, R., Wang, X., Bachmann, M., and Schlaepfer, D., "Correction of cirrus effects in Sentinel-2 type of imagery", Int. J. Remote Sensing, Vol.32, 2931-2941 (2011).

Kaufman, Y., Sendra, C. Algorithm for automatic atmospheric corrections to visible and near-IR satellite imagery, International Journal of Remote Sensing, Volume 9, Issue 8, 1357-1381 (1988).

Schläpfer, D. et al., "Atmospheric precorrected differential absorption technique to retrieve columnar water vapour", Remote Sens. Environ., Vol. 65, 353366 (1998).

Lecture Notes

RS Image Corrections and Registration Radiometric and Geometric Corrections

Raul Q. FeitosaGilson. A. O. P. Costa

Patrick N. Happ

Documents

RS Image Corrections and Registrationraul/RS/ELE2769_03_RSImage... · G R Y RE NIR1 NIR2 Digital Sensors. Radiometric Correction •Digital numbers (pixel intensity values) are unique