windows tend to enhance edges in all directions. A particular directional trend or
orientation of edges and linear features may also be enhanced by using a rectangular
filtering window.
In the K-averaging filter process (Ghandeharian et al. 2009), the filtering window is divided into sub-windows, thus creating a two-dimensional filter. A comparison of the average values and the standard deviation of the various
sub-windows determine where in the window rapid brightness variations,
e.g. ‘edges’, occur. This feature is then enhanced by averaging only those
sub-windows along the features where rapid brightness variations were detected.
Linear features can also be analyzed through the fast Fourier transformed (FFT)
of an image. The FFT can be used as an edge enhancement tool by smoothing the
transformed image by means of an average filter, and then removing the very low
and very high frequency components of the image by intensity level slicing. After
the FFT, averaging and level slicing procedure the image is inverse Fourier
transformed back into a residual image of the original scene.
14.5.3 Contrast Stretching
Synthetic aperture radar data potentially have a very wide dynamic range,
encompassing intensity levels from 0 to more than 10
7 . Once processed and output
on a CCT, the signal intensity is compressed to intensity levels which in the case of
8-bit data range from zero, for dark signatures, to 255, for very bright signatures.
Yet, the full recording range of 256 intensity levels is rarely utilized. On average,
SAR scene intensity occupies only half the digital levels available. This results in a
greatly reduced image contrast, which may potentially cover the remaining intensity levels. The intention is to make optimal use of the limited dynamic range of the
digital tape and avoid saturation of the intensity values.
Therefore, the idea behind contract stretching is to redistribute the pixel values
utilizing the full brightness range of the digital tape. A number of algorithms have
been developed to enhance digital imagery in this fashion. Contrast stretching can
also be applied to match image intensities to the characteristics of display and
recording devices. The data manipulation process usually consists of two steps;
First, the distribution, or histogram, of the pixel intensities in a given scene within
the potential range, e.g. zero to 255, is evaluated. Then, the pixels are redistributed
within the range of 0–255. Criteria for the redistribution can be selected by the
image analyst and may include linear contrast stretch and non-linear contrast
stretch. It is important to note that digital contrast enhancement should only be
performed after other processing procedures have been completed, since contrast
stretching results in some manipulation of the original pixel values.
The simplest form of contrast enhancement is called linear contrast stretching. It
is based on the following procedure: Once the histogram of scene intensity values is
generated, a digital number (DN) in the low range of this histogram is assigned to
very dark, e.g. zero. Likewise, a high DN value in the upper range is assigned to
292
S. Pirasteh et al.
orientation of edges and linear features may also be enhanced by using a rectangular
filtering window.
In the K-averaging filter process (Ghandeharian et al. 2009), the filtering window is divided into sub-windows, thus creating a two-dimensional filter. A comparison of the average values and the standard deviation of the various
sub-windows determine where in the window rapid brightness variations,
e.g. ‘edges’, occur. This feature is then enhanced by averaging only those
sub-windows along the features where rapid brightness variations were detected.
Linear features can also be analyzed through the fast Fourier transformed (FFT)
of an image. The FFT can be used as an edge enhancement tool by smoothing the
transformed image by means of an average filter, and then removing the very low
and very high frequency components of the image by intensity level slicing. After
the FFT, averaging and level slicing procedure the image is inverse Fourier
transformed back into a residual image of the original scene.
14.5.3 Contrast Stretching
Synthetic aperture radar data potentially have a very wide dynamic range,
encompassing intensity levels from 0 to more than 10
7 . Once processed and output
on a CCT, the signal intensity is compressed to intensity levels which in the case of
8-bit data range from zero, for dark signatures, to 255, for very bright signatures.
Yet, the full recording range of 256 intensity levels is rarely utilized. On average,
SAR scene intensity occupies only half the digital levels available. This results in a
greatly reduced image contrast, which may potentially cover the remaining intensity levels. The intention is to make optimal use of the limited dynamic range of the
digital tape and avoid saturation of the intensity values.
Therefore, the idea behind contract stretching is to redistribute the pixel values
utilizing the full brightness range of the digital tape. A number of algorithms have
been developed to enhance digital imagery in this fashion. Contrast stretching can
also be applied to match image intensities to the characteristics of display and
recording devices. The data manipulation process usually consists of two steps;
First, the distribution, or histogram, of the pixel intensities in a given scene within
the potential range, e.g. zero to 255, is evaluated. Then, the pixels are redistributed
within the range of 0–255. Criteria for the redistribution can be selected by the
image analyst and may include linear contrast stretch and non-linear contrast
stretch. It is important to note that digital contrast enhancement should only be
performed after other processing procedures have been completed, since contrast
stretching results in some manipulation of the original pixel values.
The simplest form of contrast enhancement is called linear contrast stretching. It
is based on the following procedure: Once the histogram of scene intensity values is
generated, a digital number (DN) in the low range of this histogram is assigned to
very dark, e.g. zero. Likewise, a high DN value in the upper range is assigned to
292
S. Pirasteh et al.