Skip to content

Latest commit

 

History

History
83 lines (46 loc) · 4.51 KB

File metadata and controls

83 lines (46 loc) · 4.51 KB

Speckle reduction of SAR images

// Kristian Aalling Sørensen - 2017.

Content:

  • Introdution to speckle
  • Implemented fitlers
  • Description and illustration
  • Link to the original report with code, background, results and analyses.

Speckle noise inherent in SAR images weakens the meaningful applications of the SAR. The reduction of the speckle noise is implemented, leading to actual filtering algorithms, based on the early propositions for speckle filtering. To evaluate the performance of the different algorithms, several parameters are introduced. All parameters are based on statistical theory and should be well known within the field. The only exception is the EQP developed by the author to quantify the quality of the edges after a speckle filtering(using an edge detection).

In addition to the parameters, the analysis builds on the visual interpretation of the SAR images. Conclusively, the Lee-filter turns out to be superior to the others in terms of speckle reduction and edge preservation, having only a slight disadvantage in dealing with close targets.

Here, filters useful for all types of SAR images are devloped. The fitlers are not assuming that the SAR images are e.g. Sentinel-1. The filters themselves calculate all the neccesary statistics etc. to fully function.

The report was made on my second year of University, and I apologize for the quality of the plot. Still, I hope it is useful.

MATLAB-Speckle-reduction

Severeal different speckle filters are implemented using Matlab to reduce the noise for different types of applications. The filters are not implemented using Matlab inherent functions (like e.g. its kernel function), and can therefore be directly translated to other languages like Python.

Implementations:

The following filters have been developed:

-Mean filter.

-Frost filter.

-Modified frost filter(added thresholds).

-Lee filter(Using Masks, ENL ect).

Note these fitlers are developed without the use of loops in Matlab.

Furhter, a simple sobel edge detection algorithm is implemented ledning to and edge quality paramter.

Description:

SAR images have a phase and amplitude. Below, a Sentinel-1 SAR VV GRD IW image aquired over Denmark is displyed. 6 Different regions are highlighted.

Alt text

Due to the inherent speckle in SAR images, speckle reductions must often times be computed. Speckle reduciton can (in general) only be performed on homogenous regions, such as e.g. region 1, region 2 or region 3. If edges occur in images, adaptive speckle reduction must be perfomed as to perserve the edges. Below, a speckle filtering has been done on region 5 (two fields). To the left, we see the origial image being "granular" - this is speckle. We then then see the different filters implemented (the implementation are shown in frost.m etc.). The box filter performs poorly at the edges.

Alt text

Visually, the Lee filter outperforms the others. Below, we can more clearly see how the Lee filter outperform the other fitlers, both with spckle filtering and with edge filtering. It is here clear to see the edge is sharper for the Lee filter, while the homogenous region is speckle filtered.

Alt text

Inhomogeneous areas

Below, a speckle filtering and edge detection has been performed on an inhomogenuous region (region 6, a large city), with different theresholds. For information on the thresholds, see the report below.

Alt text

It is clear to see that the speckle filtering performs poorly on inhomogenous regions. In fact, this is due to the so-called "Equivalent Number of Looks" (ENL). For the Sentinel-1 IW GRP Images (used here), the images have been averaged on average 4.4 times. This statistics is in fact computable from the different regions for both polarisations, as shown below. Here, we see that region 1,2 and 3(homogeneous) have statistics comparable to the true ENL. The inhomogenuous does not.

Alt text

In the report "Speckle reduction of SAR images", we saw why the Lee filter outperformed the other filters. Still, a trade-off must be made as this is computationally more expensive than the others, see the figure below:

Alt text

Report with code

For motivation or information of the filters, see:
https://drive.google.com/file/d/1UXcGZB0k54sjeEiX6oCoyUhCtHtZ0CMQ/view?usp=sharing