Back to Radar Courses
39. Image interpretation: Tone
Different surface features exhibit different scattering characteristics:
Urban areas: very strong backscatter
Forest: medium backscatter
Calm water: smooth surface, low backscatter
Rough sea: increased backscatter due to wind and current effects
The average backscattering coefficient σ 0 differs according to the types of surface, so it may be reasonably expected that surfaces with different values of σ 0 would produce different grey levels in the image.
40. Image interpretation: Speckle
A detailed analysis of the radar image shows that even for a single surface type, important grey level variations may occur between adjacent resolution cells. These variations create a grainy texture, characteristic of radar images. This effect, caused by the coherent radiation used by radar systems, is called speckle. It happens because each resolution cell associated with an extended target contains several scattering centres whose elementary returns, by positive or negative interference, originate light or dark image brightness.
This creates a "salt and pepper" appearance.
An example of speckle is shown in the figure.
The SAR scene acquired on 21/4/1994 over Tiber Valley (I) shows some agricultural fields located along the Tiber River north to Rome, in the central part of Italy.
The homogeneous patches representing the fields have high variability in backscattering due to the speckle noise. This results in a grainy image, which renders difficult the interpretation of the main features of the surface imaged by the SAR.
Speckle is a system phenomenon and is not the result of spatial variation of average reflectivity of the radar illuminated surface. For a high resolution radar, there may be useful scene texture which differs from the speckle.
This is the case for example in forested zones where the combined effects of radar illumination and tree shadowing create a rougher texture granularity than the speckle. In this case, there exists a spatial variability of the physical reflectivity of the illuminated zone. In a radar image we may find:
- zones where the only image texture is related to speckle that we may call regions "without texture" (extended homogeneous target)
- zones "with texture" that have spatial variations in scene reflectivity in addition to speckle
Thus, in the case of "no texture" zones, it becomes possible to study the statistical distribution of the backscattered radar signal, which helps to estimate certain radar characteristics.
Speckle can be reduced by two methods:
SAR image multi-look processing
Independent measurements of the same target can be averaged in order to smooth out the speckle. Actually, it is obtained by splitting the synthetic aperture into smaller sub-apertures, the so called "looks", each separately processed and then averaged.
The different looks are averaged to reduce the grey level random variations provoked by speckle. For N statistically independent (non-overlapping) data sectors, the speckle variance is reduced by a factor of N. Likewise, the resolution is degraded by a factor of N.
In such a way, we can for example have 8-look images. A compromise has to be found between desired spatial resolution and an acceptable level of speckle.
When the finest resolution is required, moving window filters are used. A moving window filter changes the intensity of the central pixel depending on the intensities of all the pixels within the window. Different algorithms have been proposed to properly shape the impulse response of the filter within the window.
In both cases speckle is reduced at the expenses of the spatial resolution.
41. Image interpretation: Speckle filters
We can consider that an extended area is characterised by:
- the radar backscattering coefficient σ ° (radiometric information)
- spatial variability (textural information)
The presence of speckle reduces the separability of the various land use classes, based on radiometry and texture. It is thus important to treat speckle so as to improve the possibility of separation, but with minimal loss of information.
If we take the example of homogeneous areas such as agricultural zones with a defined field pattern, the filters to be used must preserve the average backscattering value, and maintain sharp edges between adjacent fields.
In the case of textured areas, such as forest for example, the filters should also preserve the spatial variability (textural information) relating to the scene.
The figure provides an example of simple filter. On the left part of the screen is displayed the speckled SAR image, and on the right part a filtered version of the same image. The filter applied in this case is the Mean filter which substitutes the central pixel with the mean value within the window. It results in a less noisy, but blurred, image.
More complex filters try to make some hypotheses on the noise statistic. The Frost filter models the signal as a slow varying function representing the texture of the scene multiplied by a gaussian distributed noise. The result of the filtering is shown on the right part of the image, while on the left part it is possible to observe the original SAR image. Filtering with this technique results in a less noisy image, better preserving the actual texture of the scene.
A more refined technique of speckle filtering is performed with the MAP Gamma filter. This filter models the distribution of the texture as a Gamma distribution and applies a maximum likelihood estimation to retrieve the detected intensity.
A separate routine allows to detect edges and lines. The result of the filtering is showed on the right part of the figure, together with the unfiltered SAR image (left side), and it is particularly useful for classification purposes.