Digital Image Processing

Image Rectification and

Restoration

Presented by
Mr.Prafulla Kumar panda
Assistant Professor
Department of civil Engineering
Contact-9438269572
Mail: prafullapanda@cutm.ac.in
 Distorted Image

Radiometric Correction
Geometric Rectification
Grid1 to Grid2
Cord1 to Coord2
No proj to Proj

Resampling (new pixel values)

Corrected Image
(Raw data)

Enhancement Transformation Ratioing

Classification Neural Networks Modelling (GIS),

Others

Overview of DIP
 Introduction

The common image processing functions available in image

analysis systems can be categorized as follows:
– Preprocessing (on distorted images)
– Processing (on corrected images)
* Image Enhancement
* Image Transformation
* Image Classification and Analysis
 Pre-processing
• Operations that are normally required prior to the main
data analysis and extraction of information
– Radiometric corrections.
– Geometric corrections.
• Radiometric corrections
– correcting the data for sensor irregularities and
unwanted sensor or atmospheric noise, and
converting the data so they accurately represent the
reflected or emitted radiation measured by the sensor.
• Geometric corrections
– include correcting for geometric distortions due to
sensor-Earth geometry variations, and conversion of
the data to real world coordinates (e.g. latitude and
longitude) on the Earth's surface.
 Radiometric Distortion
• What is a radiometric distortion?
• It’s an error that influences the radiance or radiometric
value of a scene element (pixel).
Why?
– Signal traveling through atmosphere; atmosphere affects
the signal
– Sun illumination influences radiometric values
– Seasonal changes affect radiometric values
– Sensor failures or system noise affects values
– Terrain influences radiance
 Radiometric Distortion

Radiometric distortions are due to:

• Sensor noise and failures
• Seasonal variations
• Atmospheric effects
• Terrain
 Radiometric Correction
• Radiometric correction is used to modify DN values in
order to account for noise, i.e. contributions to the DN
that are a function NOT of the feature being sensed but
of the atmosphere or the sensor itself.

– The purpose is to ensure accuracy in the relationship

between radiant flux leaving a surface and radiant
flux recorded by a sensor (DN).
– Internal radiometric errors include sensor
malfunction and improper calibration.
– External radiometric errors include atmospheric
effects (“fixed” with atmospheric correction).
 Sensor Failure & Calibration
• Sensor Malfunction 
– Sensors sometimes simply do not operate correctly,
resulting in radiometric error.
• Common forms of this type of error include:
– line drops,
– striping (banding),
– line-start error
– random noise or spikes
Missing data due to sensor failure results in a line of 0 values
every 16th line for TM data (or 6th line for MSS) .. since
there are 16 sensors for each band, scanning 16 lines at a
time.
• Data can also show excessive speckle 'salt and pepper'
effect (high and low values); this can be corrected using a
box filter (3 x 3, 5 x 5): mean, median or modal.
 Sensor Noise and Failure

UKZN August 09 ENVS720

 Line drop out
Line drops are caused by a detector failing to operate during a
scan

A serious problem, because data never collected cannot be

restored.

However, we can improve our ability to interpret the data by

estimating DNs for each pixel on the bad line.
Correction method:
1. Identify defective scan lines by comparing scene
average DN to average DN of each scan line
2. Replace pixel in defective scan lines by the average of
the two pixels directly above and beneath the pixel of
the defected scan line
 Correcting Sensor Errors

• Line drops can be corrected using the formula:

 DNi  1, j , k  DNi  1, j , k 
DNijk  Int 
 2 
– where DNijk = output DN;
– DNi-1,j,k = preceding line pixel DN; and
– DNi+1,j,k = succeeding line pixel DN.
Line dropout
Line Start errors are caused by a detector failing to operate
during a scan.
Similar to line drop.

Data never collected

cannot be restored.

However, DNs for each

pixel on the bad line
can be estimated, as
was done to correct
line drop out.

If you do not want to

calculate values for the
entire affected line, you
must intervene more B. Rundquist
directly.
Striping is caused by a detector that is out of adjustment.
That is, one sensor is not
responding like the others.

Does not represent a

complete loss of data.

The computer first

calculates which lines are
striped by calculating a
mean for each 6th line
(detector).

The means should match

closely. If there is
significant differences, the
detector is striping the
image.
The mean for that detector
B. Rundquist
can then be adjusted to
ENVS720
match the others.
 Striping

UKZN August 09 ENVS720

 Random Noise or Spikes

• Cause: Transmission errors or temporary disturbances

• Correction method:
– 1. Detect spikey by comparing DN with DN of its
surrounding pixels (neighbors')
– 2. Replace DN with DN value interpolated from the
surrounding pixels
UKZN August 09 ENVS720
 ENVS720

UKZN August 09
 Atmospheric Effects
• Absorption
– Water vapour
– Carbon dioxide
– Ozone
 Atmospheric Interference

HAZE
• Lower wavelength bands are increasingly subject to haze,
which falsely increases the DN value.
• This needs correction in some cases, for example to mosaic
scenes with different amounts of haze, or to generate band
ratios, where the resultant values may be affected.
• The effect of haze diminishes with increasing wavelength, but
clouds affect all visible and IR bands, hiding features twice:
once with the cloud, once with its shadow. Only in the
microwave, can energy penetrate through clouds.
 Atmospheric Effects

UKZN August 09 ENVS720

 Haze Correction
• The simplest method is known as dark object subtraction
• Identify black bodies: clear water and shadow zones with zero
reflectance in the infrared bands
– Identify DN values at shorter wavelength bands of the
same pixel positions. These DN are entirely due to haze
– Subtract the minimum of the DN values related to black
bodies of a particular band from all the pixel values of that
band
UKZN August 09 ENVS720
• Corrections are needed to compensate for sun illumination
variations resulting from different scene acquisition dates
• Applications
– Change detection studies
– Mosaics
 Summary
• We need to correct for atmospheric effects, variations in sun
illumination and system defects
• Atmospheric correction is the most complex correction
• In many cases only haze correction is applied (dark object
subtraction)
• If we want to relate RS data with field spectrometric
measurements we need to apply full atmospheric corrections,
taking into account wavelength dependent factors, such as
diffuse sky irradiance, transmission coefficients, path radiance
and sensor calibration coefficients
 Geometric Distortions & Restoration

• Any remotely sensing image, regardless of whether it is

acquired by a multispectral scanner on board a satellite, a
photographic system in an aircraft, or any other
platform/sensor combination, will have various geometric
distortions.
• This problem is inherent in remote sensing, as we attempt
to accurately represent the three-dimensional surface of
the Earth as a two-dimensional image.
• All remote sensing images are subject to some form of
geometric distortions, depending on the manner in which
the data are acquired.
 Sources of Distortion
• the perspective of the sensor optics,
• the motion of the scanning system,
• the motion and (in)stability of the platform,
• the platform altitude, attitude, and velocity- causes
changes in pixel recording- roll, pitch, yaw
• the terrain relief, and
• the curvature and rotation of the Earth.
• Tilt- a slight tilt of a satellite can have enormous
distortions since it is positioned 100’s of km from the
earth
 Exterior Orientation

– attitude (roll, pitch, yaw)

– three dimensional position (x, y, z)
 Geometric Distortion (Pushbroom Sensor)
due to platform exterior orientation changes

(a) (b) (c)

(a) Altitude change [
planimetric position (x,y)
changes cause directly
corresponding changes in
the location of sensed data ]
(b) Pitch change
(c) Yaw change
(d) Roll change

(d)
UKZN August 09 ENVS720
 Geometric Distortion
due to platform velocity changes
Roll Pitch Yaw

oversampling

undersampling

• forward platform velocity decreases (higher orbit) -> spacing between

scan lines in the along track direction decreases, i.e. oversampling
(redundancy of image data).

• forward platform velocity increases (lower orbit) -> spacing between the
scan lines in the along track direction increases, i.e. undersampling (gaps).
UKZN August 09 ENVS720
 Sources of Distortion
• Relief displacement on aerial photographs - objects that are directly
below the lens will have only their top part imaged where as other objects
will lean away from the centre of the photographs and their sides appear.
• Images from across-track scanning systems are distorted due to the
rotation of the scanning optics. As the sensor scans across each line, the
distance from the sensor to the ground increases further away from the
centre of the swath.
• Skew distortion - caused by the eastward rotation of the earth which
causes a sweep of scanning systems to cover an area slightly to the west of
each previous scan
 Surface Relief
In scanner imagery the relief
distortion occurs only in the across
Relief track direction, and its visual effect is
H such that the relief appears to be
displaced away from nadir along
scan.

d x
d = relief distortion (m) Example: MODIS over Mount
Everest (8.84km high). The mountain
d=hx/H–h top may be located 0 - 1170km (edge
of scanline) from nadir. The
H = Sensing altitude relative to horizontal datum (m) corresponding relief distortion is 0 -
h = Relief height relative to horizontal datum (m) 14.8 km.
X = Distance of relief from nadir (m)
 Image Rectification & Restoration

• Distortions can be classified into systematic and non-

systematic /random distortions
• Systematic distortions are
• Scan skew due to forward movement of the platform
• Changes in velocity of scanner
• Scanning at an angle
– These distortions are removed during pre-processing by
applying formulas derived by modeling the sources of
distortion mathematically.
• Non-systematic distortions
– Variation in platform altitude, relief displacement, tilt, velocity,
earth rotation
• Distortions vary from area to area
• Corrections are done using ground control points distributed on
the image e.g. Highway intersections, river confluences
• GCPs are located both in terms of their two image coordinates
(column, row numbers) on the distorted image and in terms of
their ground coordinates.
• The values are submitted to a regression to determine
coefficients for two coordinate transformation equations, which
are then used to precisely estimate the coordinates of the
distorted image for any map position.
• Transformation techniques such as Affine or conformal are used
to correct for the distorted images
 Problems with GCPs
1) Availability not guaranteed
• over repetitive/unstructured surfaces e.g., dense forest, desert, ocean
• because of occlusion by clouds, fog, cloud/relief shadows

2) Sampling ?
• the number and spatial distribution of GCPs that should be collected
is poorly defined
• this usually leads to locally redundant (i.e. excessive effort) or
insufficient (i.e. inaccurate) use of GCPs

3) Accuracy
• GCP positional accuracies vary as a function of numerous variables -
particularly if they are collected manually
• After transformation, the original arrangement of
pixels is distorted
• Grid centers from the map-registered pixels grid
will not usually project to exact pixel centre
locations in the image
• Resampling is therefore necessary in order to
produce a regular grid size of pixels
 Interpolation Techniques

• Nearest neighbour – simply chooses the actual pixel that has

its centre nearest the point located in the image. This pixel is
then transferred to the corresponding display grid location.
– Advantage – computationally simple to implement and
original values of the image are not altered thus ideal
when image is used for classification
– Disadvantage – Features in the output matrix maybe offset
by up to one half pixel
 Nearest Neighbour

Bilinear

UKZN August 09 ENVS720

 Nearest Neighbour

Bilinear

UKZN August 09
• Bilinear Interpolation – Output value is calculated by
finding the weighted average value from the 4
surrounding pixels in the original image. Resulting image is
smooth without offsets. But the process alters the gray
levels of the original image.
• Cubic convolution – Takes into consideration all 16
surrounding pixels for computing the output value
– Adv – it avoids the disjointed appearance of the nearest
neighbour method & provides a sharper image than the bilinear
method.
– But the technique also alters the original image gray levels.