摘要(英) |
Speckle, appearing in Synthetic Aperture Radar (SAR) image as granular noise of mixed bright and dark pixels, is due to coherent interference of backscattered wavelets from scatters with their phases randomly distributed in a resolution cell. Speckle in SAR image complicates the image interpretation problem by reducing the effectiveness of image segmentation and classification. To alleviate the speckle effect, many algorithms have been devised to suppress speckle. Among them, the Refined Lee filter which was developed based on Rayleigh probability model has received widespread acceptance, because of its excellent characteristics of smoothing speckle noise while preserving edges, line features and image resolution. Another frequent applied algorithm is the sigma filter. The sigma filter was developed based on the two sigma range of Gaussian distribution. However, its deficiencies of introducing bias and blurring bright targets make it less desirable ,especially for single–look SAR data. In this thesis, an improved sigma filter was developed, and our test results show that its effectiveness is comparable to the Refined Lee filter.
This research is derived from the basic concept of sigma filter. The sigma filter is known to have the following deficiencies:(1) the estimated amplitudes or intensities are biased , because , unlike Gaussian distribution, the speckle distribution is not symmetrical about its mean, (2) many isolated dark pixels remain not filtered, because of their small sigma range that excludes other brighter pixels, (3) bright point targets are smeared due to the large sigma range that includes all pixels. To compensate for these deficiencies, we devised an improved sigma filter. We divided the development process into two stages. In the first stage, sigma intervals were recomputed base on SAR speckle distributions to maintain their mean values in the filtered image, and different ways to estimate the initial mean value were implemented. The first two deficiencies were successfully eliminated, but bright edges and targets were blurred due to losing image resolution, So in the second stage, we incorporated the MMSE (Minimum Mean Square Error) Method into the sigma filter to overcome the losing image resolution problem. In addition, to preserve the bright targets, a threshold was established to retain bright areas containing more than three pixels.
The improved Sigma filter successfully compensated for the deficiencies of the original sigma filter. For illustration, the improved sigma filter was tested and evaluated using SAR data from ALOS/PALSAR and JPL/AIRSAR. Reasonably good results of speckle reduction and edge preserving were obtained. |
參考文獻 |
[1] J . W . Goodman , “ Some fundamental properties of speckle , ” J .Opt . Sco . Am , 66 (11),p1145-1150,1976.
[2]J.S.Lee,et al.“Speckle Filtering of SAR images:A Reviews,"Remote Sensig Reviews , Vol.8,PP,313-340,1994.
[3]J.S.Lee,“Digital image enhancement and noise filtering by use of local statistics,” IEEE Trans . PAMI ,Vol.2 No.2,1980.
[4]D.T.Kwan,et al.“Adaptive Noise filtering for Images with Signal dependent noise,” IEEE Trans.PAMI, Vol.7,No.2,1985.
[5]A.Lopes,et al.“Adaptive speckle filters and scene homogeneity,”IEEE TGARS,Vol.28,No.6,1990.
[6]J.S.Lee,“ Refined filtering of image noise using local statistics,“ CVGIP , vol .15,p380-389,1981.
[7]J.S.Lee,“Digtal image noise smoothing and the sigma filter,” CVGIP, Vol. 24, p255-269,1983
[8]J.S.Lee,“A Simple speckle smoothing algorithm for SAR images,”IEEE Trans . SMC ,Vol.13 No.1,1983.
[9]J.S.Lee,K.Hoppel and S.A.Mango,“Unsupervised Estimation of Speckle Noise in Radar images,” Int.J.of Imag.Sys. and Tech., vol.4,p298-305,1992.
[10]Ulaby F.,Moore,R. and A. Fung, Microwave Remote Sensing, Vol. 3, Artec House,Norwood,MA,1986. |