Overview of Image Processing
S3
2.3
Image Distortion and Rectification
Obviously, the ideal image is an accurate reproduction of ground reflectance
in the various spectral bands. However, there are various factors that make
this impossible. Thus, most recorded images suffer from various degrees of
distortion. There are two main types of distortion, radiometric and geometric.
The former is the inaccurate translation of ground reflectance to gray values,
typically arising due to atmospheric interference and/or sensor defects. The
latter refers to shape and scale distortions owing usually to errors due to
perspective projection and instrumentation. We provide an overview of these
distortions along with schemes used to rectify the errors.
2.3.1
Radiometric Distortion
This refers to inaccurate representation of relative brightness levels (or intensity
values) either in a given band across pixels or across bands for a given pixel. The
ideal measured brightness by a sensor at a pixel in a given band is proportional
to the product of the surface reflectance and the sun's spectral irradiance
assuming uniform surface reflectance over the area contributing to the pixel.
Thus the ideal image captures the relative brightness as measured at the ground.
However, the intervening atmosphere between the ground and a space based
sensor causes inaccuracies in the recording of the relative levels.
Atmosphere induced radiometric distortion is due primarily to scattering of
electromagnetic radiation by either the air molecules or suspended particles.
Part of the down -going radiation from the sun is scattered back and away from
the earth by the atmosphere itself. Also, apart from the radiation reflected by
the area covered by the pixel of interest, radiation reflected from neighboring
regions is scattered by the atmosphere into the pixel of interest. The main effect
of both types of scattering is loss of image detail. It also causes inaccurate
representation of radiometric levels across the bands due to the wavelength
dependent nature of scattering.
Correction of atmospheric distortion can be done in either a detailed or
gross fashion. Detailed correction requires information regarding the incident
angle of the sunlight and atmospheric constituents at the time of imaging. The
scattering effects are then modeled into atmospheric attenuation as a function
of path length and wavelength. There are a number of atmospheric models
that allow the conversion of at-sensor image data to transform into ground
reflectance. Some of the methods are flat field correction, internal average
relative reflectance and empirical line method (Rast et al. 1991; Smith and
Milton 1999). Because detailed atmospheric information is hard to obtain, gross
or bulk corrections are more common. Bulk brightness correction attempts
mainly to compensate for at-sensor radiance or path radiance. Path radiance
effect for a given pixel refers to interfering scatter from the atmosphere as well
as from other pixels. A typical approach, referred to as dark object subtraction,
is to subtract the lowest gray value of each band from the values of other pixels
S3
2.3
Image Distortion and Rectification
Obviously, the ideal image is an accurate reproduction of ground reflectance
in the various spectral bands. However, there are various factors that make
this impossible. Thus, most recorded images suffer from various degrees of
distortion. There are two main types of distortion, radiometric and geometric.
The former is the inaccurate translation of ground reflectance to gray values,
typically arising due to atmospheric interference and/or sensor defects. The
latter refers to shape and scale distortions owing usually to errors due to
perspective projection and instrumentation. We provide an overview of these
distortions along with schemes used to rectify the errors.
2.3.1
Radiometric Distortion
This refers to inaccurate representation of relative brightness levels (or intensity
values) either in a given band across pixels or across bands for a given pixel. The
ideal measured brightness by a sensor at a pixel in a given band is proportional
to the product of the surface reflectance and the sun's spectral irradiance
assuming uniform surface reflectance over the area contributing to the pixel.
Thus the ideal image captures the relative brightness as measured at the ground.
However, the intervening atmosphere between the ground and a space based
sensor causes inaccuracies in the recording of the relative levels.
Atmosphere induced radiometric distortion is due primarily to scattering of
electromagnetic radiation by either the air molecules or suspended particles.
Part of the down -going radiation from the sun is scattered back and away from
the earth by the atmosphere itself. Also, apart from the radiation reflected by
the area covered by the pixel of interest, radiation reflected from neighboring
regions is scattered by the atmosphere into the pixel of interest. The main effect
of both types of scattering is loss of image detail. It also causes inaccurate
representation of radiometric levels across the bands due to the wavelength
dependent nature of scattering.
Correction of atmospheric distortion can be done in either a detailed or
gross fashion. Detailed correction requires information regarding the incident
angle of the sunlight and atmospheric constituents at the time of imaging. The
scattering effects are then modeled into atmospheric attenuation as a function
of path length and wavelength. There are a number of atmospheric models
that allow the conversion of at-sensor image data to transform into ground
reflectance. Some of the methods are flat field correction, internal average
relative reflectance and empirical line method (Rast et al. 1991; Smith and
Milton 1999). Because detailed atmospheric information is hard to obtain, gross
or bulk corrections are more common. Bulk brightness correction attempts
mainly to compensate for at-sensor radiance or path radiance. Path radiance
effect for a given pixel refers to interfering scatter from the atmosphere as well
as from other pixels. A typical approach, referred to as dark object subtraction,
is to subtract the lowest gray value of each band from the values of other pixels