Preprocessing of Remotely Sensed Data

Geometric Correction

• L ECTURE 5

• MUHAMMAD HASAN ALI BAIG

 Introduction
o Preprocessing
o Remotely sensed data generally contains flaws or deficiencies with respect to particular applications
o The correction of deficiencies and removal of flaws present in data is termed as preprocessing
o Preprocessing of data is carried out before the data is used for a particular purpose
o Required preprocessing operations depends upon the intended use of data
o Preprocessing Operations
o Geometric Corrections
o Radiometric Corrections
o Atmospheric Corrections

o Guidelines
o All of these operations are not required in all cases
o The analyst should decide appropriate preprocessing techniques on the basis of the nature of information to be extracted from
remotely sensed data

o Data errors such as missing scanlines can’t be removed, rather replaced with some other data that are felt to be
better estimates of the true unknown values
Missing Scanlines
 Missing Scan Lines
o Digital image contain partial / complete missing scan
lines
o Errors in scanning / sampling equipment
o During transmission / recording

o Observed during
o Visual examination
o Interpretation
o Statistical calculations
o Appear as black or white lines
o For 8 bit image black means 0 value while 255 for white
o Missing scan lines means no data has been captured /
available
o Missing or dead scanlines / stripping & banding
 Stripping – Detector to Detector Miss-calibration

1
3

o Several bright lines cross horizontally the

image

2 o They are spaced of about 15 pixels

o It seems that one detector among 16 has not
been calibrated correctly
 Banding – Scan to Scan Miss-calibration

o Scan to Scan mis-calibration errors occur when calibration algorithms treat forward and reverse scans
separately so that one direction of scan has a different bias than the other
o Scan to Scan miss-calibration may be visible in data from older instruments prior to Landsat 7
 Missing Scan Lines
o Push broom-type imaging instruments are designed to acquire complete rows of images
using a linear array of detectors
o Remote sensing images often suffer from the common problems of stripe noises and linear
or random dead pixels
o Estimation of missing data / values is done
o Auto correlation is the relation of one value with the neighboring value (s)
o De-Stripping
o The correction of image stripes is commonly called as image de-striping
 Methods for Calculating Missing Scan Lines
o Method 1
o Simplest method is replacing dropped / missing scan line with the value of preceding line
o Vij = Vij -1

o Method 2
o Value of missing line is calculated as average sum of above and below lines
o Vij = (Vij -1 + Vij +1)/2

o Method 3
o In multispectral imagery if the missing pixels are making up of two other bands then these are correlated on
a pair by pair basis
o High correlations are found for bands in the same region of the spectrum
o Vi,j,k = M[Vi,j,r - (Vij +1,r + Vij-1,r)/2]+ (Vi,j +1,k + Vi,j-1,k)/2
The symbol M is the ratio of the standard deviation of the pixel values in band k and the standard deviation
of pixel values in band r
 Linear Method for De-Stripping
o Uses linear expression between input and output values
o Assumes that all detectors sees similar distribution of landcover categories that are present in image area and
therefore, produces identical histogram for n detectors
o Hence means and standard deviations of the data measured by each detector must be same
o To eliminate the stripping effects of detector imbalance, the means and standard deviation of the n
histograms are equalized, that is forced to equal a chosen value
o Usually the means of n individual histograms are made to equal the mean of all pixel in the image, and the
standard deviations of the n individual histograms are forced to be equal to the standard deviation of all pixels
in the image
o Linear method is applied pixel by pixel basis to an image, if the image size is too large then inordinately slow
response is observed
 Histogram Method for De-Stripping
o Histogram matching method is based on the assumption that the output from a detector is a linear function
of input value
o rout = offset + gain x rin (where r is the radiance and input value is the value of image be corrected)
o Histogram matching also known as histogram specification is the transformation of an image so that its
histogram matches a specified histogram
o It is used to normalize two images (reference and target), while the two images were acquired at the same
local illumination (such as shadows) over the same location, but by different sensors, atmospheric conditions
or global illumination
o Histogram matching can be used to balance detector responses
o Cumulative distribution functions / frequency histogram of complete image is calculated by carrying out class
by class summation of n individual detector histograms e.g. 6 for Landsat MSS, the histogram for detector 1 is
computed from the pixel values on scanline 1, 7, 13 and detector 2 pixel values on scanlines 2,8,14 and so on
o Any class number k in an individual histogram of the stripped image may be replaced with values of j class
histogram
Geometric Correction and Registration
 Introduction
o Information extracted from remotely sensed images
is integrated with map data in a GIS or presented as a
map like form
o Sometimes pair of images are used to develop Digital
Elevation Model (DEM)
o Map Projection is a device for representation of
curved surface (Earth) on a flat sheet of paper
o Images from different sources being combined must
be expressed in a common coordinate system
o Geometric Correction is the transformation of
remotely sensed image so that it has the scale and
projection properties of a given map projection
 Introduction
o Georectification To take an image that has not been adjusted to be in a known coordinate system, and put it
into a known coordinate system. Usually this means taking an image that is in its original geometry, and
putting it into a map projection. There are different ways to do this. Perhaps the most common way is to
identify a set of points in the image for which the latitude and longitude or map coordinates are known, and
use them to warp the image into a map projection.

o Orthorectification To take an image in its original geometry and very accurately adjust it so that it is in a
known coordinate system, with distortions due to topographic variation corrected. An orthorectified image
has uniform scale throughout the image.

o Image Registration is the fitting of the coordinate system of one image to that of second image of the same
area

o Georeferencing To take an image that is already in a known coordinate system, and provide the information
necessary for software to understand which coordinate system it is in.
 Geometric Corrections
oAllremote sensing imageries are inherently subject to
geometric distortions
oThese distortions may be due to several factors. Such as;
o The perspective of sensor optics
o The motion of scanning system
o The motion of platform
o Platform altitude and velocity
o Terrain relief
o Curvature and rotation of the earth
 Geometric Corrections
o Geometric
corrections are intended to compensate for those distortions which
make geometric representation of the imagery as close as possible to the real
world
o Many of the variations are systematic, are predictable In nature and can be
accounted for by accurate modeling of the sensor and platform motion and
the geometric relationship of the platform with the earth
o Other non systematic or random errors cannot be modeled or corrected in this
way
o Therefore, geometric registration of the imagery to a known ground
coordinate system must be performed
o The systematic and non systematic errors are discussed below
 Geometric Corrections
o Geometric
corrections are intended to compensate for those distortions which
make geometric representation of the imagery as close as possible to the real
world
o Many of the variations are systematic, are predictable In nature and can be
accounted for by accurate modeling of the sensor and platform motion and
the geometric relationship of the platform with the earth
o Other non systematic or random errors cannot be modeled or corrected in this
way
o Therefore, geometric registration of the imagery to a known ground
coordinate system must be performed
 When geometric corrections are required?
o To transform an image to match a map projection
o To locate point of interest on map and image
o To bring adjacent images into registration
o To overlay temporal sequences of images of the same area, perhaps acquired from different sensors
o To overlay images and maps with in a GIS
 Types of Geometric Errors
o Remotely sensed imagery typically exhibits internal and external geometric error
o It is important to recognize the source of the internal and external error and whether it is systematic
(predictable) or nonsystematic (random)
o Systematic geometric error is generally easier to identify and correct than random geometric error
o Internal geometric errors are introduced by the remote sensing system itself or in combination with
Earth rotation or curvature characteristics. These distortions are often systematic (predictable) e.g.
skew caused by Earth rotation effects (Landsat 7 ETM+ imagery is slightly westward displaced due to
earth’s eastward rotation
o External geometric errors are usually introduced by phenomena that vary in nature through space
and time. The most important external variables that can cause geometric error in remote sensor
data are random movements by the aircraft (or spacecraft) at the exact time of data collection,
which usually involve:
o altitude changes, and/or
o attitude changes (roll, pitch, and yaw).
 Types of Systematic Geometric Errors
o Scan Skew: it is caused by the forward motion of the platform during the time required for
each mirror sweep. The ground swap is not normal to the ground track but it is slightly
skewed, producing cross scan geometric distortion.

o Mirror Scan Velocity: The multi-spectral satellite (MSS) mirror scanning rate is usually not
constant across a given scan, producing along scan geometric distortion.

o This type of error was inherent to older sensors up till Landsat 5 and is neglect able for the
later.
 Types of Systematic Geometric Errors
o Panchromatic Distortion: The ground area imaged is proportional to the tangent of the
scan angle rather the angle itself. Since data area sampled at regular intervals, this produces
along scan distortion.

o Platform Velocity: If the speed of platform changes the ground track covered by successive
mirror scans changes producing along track scale distortion.
 Types of Systematic Geometric Errors
o Earth Rotation: The earth rotates as MSS sensors scan the terrain.
o This results in the shift of the ground swath being scanned causing along scan distortion.

o Perspective: For some applications it is desirable to have images represent the projection of
points on the earth upon a plane tangent to the earth with all projection lines normal to the
plane. This introduces along scan distortion.
 Types of Non-systematic Geometric Errors
o Altitude: If platform departs from its normal altitude, changes in scale occur.
o Attitude: One of the sensor system axes usually maintained normal to the earth’s surface
and introduces geometric distortions.
 Geometric Rectification
o The basic purpose of carrying out geometric correction is to orient the original data
collected so that it has the desired scale and map projection or simply it is made
planimetric.
o A map projection is a device for the representation of a curved surface on a flat sheet of
paper.
o Different types of map projection have been developed, however, each projection tries to
preserve only some property of the mapped area, and either it is uniform representation of
the areas or shape, or preservation of correct bearings.
o Only one such property can be correctly presented, though several projection attempts to
compromise by minimizing distortion in to two or more map characteristics.
o The angular velocity of the earth varies with latitude, so this skew becomes more
prominent at higher latitudes
 Geometric Correction Process
o Determination of a relationship between the coordinate system of map and image
o Establishment of set of points defining pixel centers in the corrected image with that, when
considered as a rectangular grid, define an image with the desired cartographic properties
and
o Estimation of pixel values to be associated with those points
o The relationship between the two coordinate systems (map and image could be defined by
two methods).
o Orbital Geometry of satellite platform
o Transformation using ground control points (GCP’s)
 Orbital Geometry Model
o Orbital Geometry methods are based on knowledge of characteristics of the orbit of the
satellite platform.
o It is a simple method of correcting the coordinate system of remotely sensed images using
approximate orbit parameters.
o This method is suitable only for use of images that have been acquired from sensors having
narrow angular field of view such as Landsat TM and SPOT HRV.
o However, it can be easily extended to deal with imagery for any similar sensor provided that
the nominal orbital parameters of satellite are known or can be estimated.
 Orbital Geometry Model
o Since the method is based on nominal orbital parameters the accuracy of the geometrically
corrected image processed by this technique is not high.
o In the orbital geometry model the corrections can be modeled in three sequential steps;
o Aspect Ratio
o Skew Correction
o Earth Rotation Correction
 Aspect Ratio
o Every sensor as per its IFoV has a ground coverage i.e. its pixel size.
o As the sensor moves along a scan line, there is a small overlap, so instead of square pixel,
now it is a rectangular pixel.
o In case of Landsat MSS, the actual size of collected data is 79m x 56m.
o Thus the shape of the pixel is to be restored back to its desired square shape.
o This correction is known as the aspect ratio correction and can be represented by
transformation matrix M1.
1.41 0.00
o 𝑀1 =
0.00 1.00
 Skew Correction
o Low altitude satellites are generally places in near North South axis.
o However, to counter the forces due to flattening of the earth and the forces of stars, it is not
possible to place the satellite in perfect North South alignment
o Landsat satellite 1 to 3 has an orbital inclination of 99.09°, whereas Landsat 4 and 5 have an
inclination of 98.2°
o The satellite heading (across scan direction at the Equator) is therefore 9.09° and 8.02°
respectively, increasing with latitude
 Skew Correction
o The skew angle Θ at latitude L is given by
−1 𝑆𝑖𝑛 𝜃𝐸
o 𝜃 = 90 − 𝑐𝑜𝑠
𝐶𝑜𝑠𝐿
o Where 𝜃𝐸 is the satellite heading at equator and 𝐿is center latitude of the image
o Given the value of coordinate system of the image can be rotated through degrees anti
clockwise so that the scan lines of the corrected image are oriented in an east-west
direction using the transformation matrix M2.
𝑐𝑜𝑠𝜃 𝑠𝑖𝑛𝜃
o 𝑀2 =
−𝑠𝑖𝑛𝜃 𝑐𝑜𝑠𝜃
 Earth Rotation Correction
o As the satellite moves southwards along its orbit, the earth rotates beneath it in an easterly
direction with a surface velocity proportional to the nadir point.
o To compute the displacement of last line in the image relative to the first line, determine
the time taken by the satellite sensor to scan the image and the eastwards velocity of the
earth.
o The time taken for the satellite sensor to scan the image can be found if the distance
travelled by the satellite and the satellite velocity are known.
o Both distance and velocity are expressed in terms of angular measure.
o If A is a point on the Earth’s surface corresponding to the center of the first scan line in the
image, and B is the corresponding point on the last scan line in the image, then the line AB
represents an arc of a circle centered at the Earth’s center.
o The angle at the Earth’s center O is given by angle AOB and can be calculated as the Earth’s
equatorial radius (R) is 6378km, so
 Earth Rotation Correction
o Further, the orbital period (P) of the satellite is know, this the angular velocity of the
satellite (ω) is equal to 2π(P x 60).
o The time (t) required to travel the angular distance AOB can be obtained by dividing
ω/Angle AOB.
o Subsequently, determine the displacement of the last scan with respect to the first scan
line.
o This is dependent on the central latitude (L) of the image.
o If the Earth’s surface velocity at latitude L is VE(L), it can be defined as:
o 𝑉𝐸 𝐿 = 𝑅 𝑐𝑜𝑠 (𝐿)𝜔𝐸
o Where R is the Earth’s radius, defined above, and is the Earth’s angular velocity.
o Since the Earth rotates once in 23 hours, 56 minutes and 4 seconds (i.e 86164 seconds),
then its angular velocity is simply 2π/86164 rad/sec pr 0.7279 rad/sec.
 Earth Rotation Correction
o Now at the given latitude (L) determine the skew angle Ѳ.
o So actual displacement is ωѲ x t x cosѲ.
o These computations are summarized by the term ask.
𝜔𝑒 cos 𝐿 cos 𝐿
𝑎𝑠𝑘 = = 0.0719
𝜔𝜃 cos 𝜃 cos 𝜃
o Where is the Earth’s angular velocity 𝜔𝑒 , is the satellites angular velocity 𝜔𝜃 (both given
above), and Ѳ and L are defined above.
o The transformation matrix M3 is:
1 𝑎𝑠𝑘
o 𝑀3 =
0 1
 Earth Rotation Correction
o The three transformations matrices M1, M2 and M3 given above are not applied separately.
o Instead, a composite transformation matrix, M, is obtained by multiplying the three
separate transformation matrices:
o M= M1M2M3
o The corrected image coordinate system is related to the raw image coordinate system by
x’ = Mx
 Geometric Rectification
o Geometric rectification is the process by which the geometry of an image is made
planimetric.
o It requires the use of GCP’s whose image coordinates in terms of rows and pixel number are
known along with the map or ground coordinates.
o It is an accurate process since each pixel is reference not only byt its image coordinates but
also by a ground coordinate system in a standard map projection, and hence known as
image to map rectification.
o Geometric rectification has to be done when accurate linear or areal measurements are
needed.
o Many a times, accurate geometrical rectification is not needed; specifically, when two or
more images are used to find out the changes in an area or region.
o Here, instead of carrying geometric rectification of all the images and then determining the
changes, the images are registered with respect to each other.
o This is known as image registration or image-to-image registration.
 Rectification vs Registration
The difference between rectification and registration is that in
rectification, the image is referenced to a map having a standard map
projection; while in registration, 2 images are referenced to each
other and do not have any map having a standard map projection.
 Map Rectification
o In order to rectify a remote sensing image to a map coordinate system, two basic
operations have to be performed
o Spatial interpolation
o Intensity interpolation
 Spatial Interpolation
o A large of number of GCPs are defined on the image and on the ground
o A polynomial is fitted to the control points using least squares criteria to model the
geometric error directly into the image without explicitly defining or identifying their source
o Generally, a 6 parameter affine transformation is sufficient to rectify a remote sensing
image as it can model for 6 kinds of distortions; that is:-
o Translation in x and y
o Scale changes in x and y
o Direction
o Skew and rotation
 Spatial Interpolation
o x ’ = a0 + a1x + a2y

o y ’ = b0 + b1x + b2y

where x and y are the ground coordinates values and the x’ and y ’ are the image coordinate
values. It may be noted that minimum of 3 GCPs can yield a solution. However, more than 3
GCPs are used to compute the coefficients; a0 , a1, a2, b0 , b1, b2
o This method involves the computation of root means square error for each GCPs

o Generally, the analyst specifies a threshold value (less than 3) for total error. If the total
error exceeds the threshold value, then the GCP having the largest individual RMS error is

deleted and a new set of coefficients of a0, a1, a2, b0, b1 and b2 are computed
 Intensity Interpolation / Resampling
o After the special is carried out, the next step to extract or identify the brightness value of
the new image from the old image
o This process is known as re-sampling
o The re-sampling process calculates the new pixel values from the original digital pixel values
in the uncorrected image
o There are 3 common methods for re-sampling;
o Nearest neighborhood
o Bilinear interpolation
o Cubic convolution.
 Nearest Neighborhood
o Nearest neighborhood re-sampling uses the digital value from the pixel in the original image
which is nearest to the new pixel location in the corrected image
o This is one of the simplest methods and does not alter the original values but may result in
some pixel values being duplicated while others are lost
o This method also tends to result in a disjointed or blocky image appearance
 Bilinear Interpolation
o Bilinear interpolation re-sampling takes a weighted average of all the 4 pixel in the original
image nearest to the pixel location
o Brightness value of the unknown pixel is equal to summation of all the 4 pixel brightness
values divided by the inverse of square of the distances of these points
 Cubic Convolution
o cubic convolution, it even goes one step further to calculate the distance weighted average
of a block of 16 pixels from the original image which surrounds the new output pixel
location
o Both Bilinear Interpolation and Cubic Convolution produce images which are sharper in
appearance and avoid the blocky appearances of the nearest neighbourhood method
THANKS