ay, 30
su
1988 with
of July 10,
Next, an anisotropic diffusion algorithm (Perona and
Makik, 1990) is used on the SAR image to minimize
noise while preserving the position and magnitude of the
edges. Anisotropic diffusion applies a four pixel operator
to each pixel in a digital image. The equation for
computing a new pixel value in the sequence is
2p *MOy SV yl Cg Vgl Co Vp tp VD (1)
J"! the new central pixel ( (t Z)th iterations) in ith
row, jth column
I. the current central pixel (ith iterations) in ith
row , jth column
à a weight value (0.25 is used)
Cy,Cg,Cg,Cy directional diffusion coefficients (a
function of VI), (N:north, S:south, W:west,
E:east)
VI a gradient (4 neighboring pixel - central pixel).
The functional form used for the directional diffusion
coefficients Cy, Cg, Cg, Cy is,
—1
lo 1 V2
ivi]
C(VI) 2| 14
(VI) É
Q)
where K = the gradient threshold.
Fig. 6 shows the resultant image after applying the
anisotropic diffusion algorithm to Fig. 3 for K=5 and
iterations=50.
Fig. 6. The resultant image after applying an anisotropic
diffusion algorithm on Fig. 3 (K=5, iterations=50).
629
Texture data are created using second order statistics in a
small-area image patch selected on the SAR image. A
local dynamic thresholding algorithm (Haverkamp et al.,
1995) was applied to both the edge enhanced image (Fig.
6) and the resultant texture data. This algorithm segments
the image into three different gray level classes.
Fig. 7 is the classified image obtained after applying the
local dynamic thresholding algorithm to Fig. 6. Fig. 8 is
the classified image obtained from the texture data.
ja NV J 7134 j^ A E
Fig. 7. The classified image segmented into three gray
level classes after applying dynamic thresholding
algorithm to Fig. 6 .
: E. : VE Fa JEDE
Fig. 8. The classified image segmented into three gray
level classes after applying the dynamic thresholding
method to gray level converted texture data.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996