Non-linear filters are especially useful when applied
on the images distorted by “salt&pepper“ noise field
(Fig. 3a). Figs 3b), 3c) and 3d) show the results after filtering
by median filter. The size of the neighborhood for
the median filter in 3b) is 3 x 3 pixels, 6 x 6 pixels for 3c)
and 10 x 10 pixels for 3d). As can be seen the median filter
suppresses the noise but slightly blurs the image (the larger
the neighborhood is, the more blurred is the output).
Figs. 3e) and 3f) are the results of filtering with adaptive
median filter. The suppression of the noise is not as
good as by median filtering but the output images are not
blurred at all. Moreover the results are almost the same for
the size of neighborhood 3 x 3 (Fig. 3e) and 10 x 10 pixels
(Fig. 3f).
The third subunit is the Frequency Domain Filtering.
Again two possibilities are available. Either filtering with
frequency filters defined in the spatial domain and then
transferred to the frequency domain by FFT (demonstration
of the Prewitt frequency filter is in Fig. 4a) or filters defined
directly in the frequency domain. Filtering in frequency
domain is useful for example when the original
image has a line-structure. That means that every even (or
every odd) line of the picture is missing (all pixels of the
line are black – Fig. 4b). Image like that can be obtained by
taking a picture of the CRT screen. The result of filtering
with gaussian frequency filter from Fig. 4c) is in Fig. 4d).
The line-structure is corrected but the output image is
slightly blurred.
The Image Processing Application’s fourth subunit is
the Noise Addition. The list of available noise fields and
their parameters with the description can be found in the
first part of this paper.