Fuzzy Methods in Nonlinear Signal Processing: Part II—Applications - Nonlinear Signal and Image Processing: Theory, Methods and Applications - page 79

FIGURE 3.7
Level-one horizontal (top), vertical (middle), and diagonal (bottom) detail signals for the Lake
Charles data set. Decomposition method: (a) biorthogonal wavelet, (b) median, (c) median affine,
and (d) scale invariant median affine. (From M. Asghar and K.E. Barner, IEEE Trans. Visualization
Comput. Graphics ,7,76-93, Mar. 2001. c 2001 IEEE. With permission.)

We compare the performance of a center affine filter with that of a WOS-
filter and an L -filter on the ISAR image* depicted in Figure 3.9a. This image

is a 128 128, 8 bits/pixel intensity image of a B-727. A pair of synthetic im-

ages, Figure 3.9b, is used to optimize each of the filters. Figure 3.10 shows
the L

×

, WOS, and center affine filtering outputs and errors. An examina-
tion of the figure shows that the WOS-filter eliminates the noise well, but
blurs details of the plane. The L -filter preserves the plane much better, but

is not very effective in removing the clutter noise. The center affine filter re-
moves the background noise to a large extent while preserving the plane and
all its details. The superior performance of the center affine filter can be ex-
plained by its affinity-based sample preference. While operating in the back-
ground noise, all observations are close to the center sample. Thus, the center

* Data provided by Victor C. Chen, Airborne Radar, Radar Division, Naval Research Laboratory,
Washington, D.C. 20375.