<< Click to Display Table of Contents >> Navigation:  »No topics above this level« Basic - Filtering - Adaptive Non-Local SAR Filtering

Purpose

 

Images obtained from coherent sensors such as SAR systems are characterized by speckle. This is a spatially random multiplicative noise due to coherent superposition of multiple backscattering sources within a SAR resolution element. In other words, speckle is a statistical fluctuation associated with the radar reflectivity of each pixel in a scene. A first contribution to reduce the speckle - at the expense of  spatial resolution - is usually performed during the multi-looking step, where range and/or azimuth resolution cells are averaged.

 

In  order to reduce the speckle noise, a non-local filter, specifically designed for SAR non-additive noise, has been developed. This tool is based on the paper reported in the References paragraph with the addition of a custom locally-adaptive similarity probability kernel. This kernel is used to overcome local distortions caused by strong backscattering variations, typical of the original filter design.

The number of iterations for the weighted denoising is internally set to 3.

Non-local-based filtering is an intrinsically, computationally demanding feature, due to its many comparisons between different image areas. As a consequence, this tool put the underlying hardware components under stress and requires a significant amount of time. For this reason, our implementation is entirely written in OpenCL to take advantage of the massive level of parallelization offered by modern graphical processing units. Hence, we warmly recommended to use this filter only in combination with a powerful GPU selected as the current OpenCL platform/device under the SARscape Common Preferences. It is always possible to select and rely to a CPU-only OpenCL platform/device, but in such a case performance will roughly be one order of magnitude lower than using a modern GPU. 

 

Technical Note

 

This tool has been tested and validated using slant-range images, for which the assumptions on the multiplicative noise probability density function are the ones reported in the referenced paper.

 

For what concerns the Estimated Number of Looks (ENL) parameter,  it is suggested to leave it as default (i.e., to -1) for single-look data, while in case of multi-looked data the ENL could be manually set by estimating it over a homogeneous area of the image.

 

The T and H parameters act as weights to balance the trade-off between the noise reduction and the fidelity of the estimate.

 

Input Files

 

Input file list

Input file names (_pwr, _rsp). This file list is mandatory.

 

Parameters - Principal Parameters

 

Average Window Size 

The size of the weighted average performed by the filter. By increasing the window size, the waving effect over homogeneous areas is reduced. However, the processing time exponentially increases in function of the window size.

 

Similarity Kernel Size 

The kernel size used to quantify the similarity probability (weight) for the weighted average.

 

Similarity Min Kernel Size 

This parameter activates the similarity probability kernel adaptivity (if this value is lower than the similarity kernel size). 

 

H factor

It modules the filter intensity. The higher the value, the higher the strength of the filter.

 

T factor

It modules the filter intensity. The higher the value, the higher the strength of the filter.

 

Equivalent Number of Looks (ENL)

The Equivalent Number of Looks is equivalent to the number of independent intensity values averaged per pixel during the multi-looking process. This parameter can be easily estimated over a homogeneous (stationary) sample in the input intensity data according to:

 

ENL =   mean2 / standard deviation2

 

If the ENL is not set (i.e., it is set to the special value -1), it is automatically estimated via an analysis over the whole image.

 

Parameters - Global

 

It brings to the general section of the Preferences parameters. Any modified value will be used and stored for further processing sessions.

 

Parameters - Other Parameters

 

It brings to the general section of the Preferences parameters. Any modified value will be used and stored for further processing sessions.

 

Output Files

 

Output file list

Output file list of the filtered data. This file list is mandatory.

 

_fil        

Filtered intensity image and associated header files (.sml, .hdr).

 

Details specific to the Units of Measure and Nomenclature of the output products can be found in the Data Format section.

 

General Functions

 

Exec

The processing step is executed.

 

Store Batch        

The processing step is stored in the batch list. The Batch Browser button allows to load the batch processing list.

 

Close        

The window will be closed.

 

Help

Specific help document section.

 

 

Specific Function(s)

 

None.

 

References

 

Deledalle, C. A., Denis, L., & Tupin, F. (2009). Iterative weighted maximum likelihood denoising with probabilistic patch-based weights. IEEE Transactions on Image Processing, 18(12), 2661-2672.