Yunham Dong
artificial unfiltered image having as the approximately same speckle level as in the filtered images. It has been noticed
that even though the filtered and the unfiltered images have the same mean and standard deviation, the appearance of
the filtered images look coarser than the unfiltered image. That is the data distribution in the co-occurrence matrix of
the filtered images is wider than that of unfiltered image. In general, among all the filters tested including the mean,
median, Lee, Kuan, Frost, and proposed filters, the median filter gives the worst result in terms of the second-order
texture restoration, which is true at least for the image used in the test. As known, the median filter assigns the median
value to the central pixel in the moving window. As a result, it has been found that, quite a few of pixels are assigned as
the same values as their adjacent pixels, causing a dramatic distortion in the co-occurrence matrix compared to the co-
occurrence matrix of the unfiltered image. As far as uniform areas are concern, the mean filter shows the best result of
restoration of the intrinsic texture. However, the mean filter also gives the poorest result in edge areas. Other adaptive
filters perform better in edge areas, but seem not as well as the mean filter does for uniform areas. In general the texture
of filtered images has been dramatically improved to a level very close to what the unfiltered image has. Details of
texture restoration of filters can be found elsewhere (Dong and Milne, 2000).
(a) Original (b) Lee
(c) Frost (d) Proposed
Figure 7. Three-dimensional view of an edge crossing area of 40 x 40 pixels. Figures shown in (a), (b), (c) and (d)
correspond to the original data, and the data processed by Lee, Frost and the proposed filters, respectively.
4 CONCLUSIONS
A speckle filtering algorithm for edge sharpening and enhancement has been proposed. Its performance in terms of edge
enhancement as well as mean preservation and standard deviation reduction has been analyzed. The preservation
(restoration) of the intrinsic texture of the filter is also tested, and found to be comparable to other adaptive filters. The
algorithm can be applied in conjunction with other speckle filters in order to sharpen edges which might be smeared.
Even for images with unsmeared edges it may be still worth implementing edge enhancement, as the response of a pulse
(a sharp edge) by SAR systems is a sinc function (a smeared edge with local oscillations) due to systems' limited
bandwidths.
94 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000.
ACI
This
REF
Any:
Tran
Canı
8, PF
Crin
Don
trans
Don:
IGA
Dou,
Belli
Fros
adap
165.
Gox
Wav
Hara
Syst
Kuai
Tran
Lee,
pp. 3
Lee,
Lee,
radai
Lee,
class
Lope
and t
Mall
Mac
Nov:
Aero
Tuke