Image Processing Reference
Comparison of optimum and median filters.
When a picture is scanned, there can be errors at the edges. An example of this is
shown in Fig. 3.9. In this case, the noisy image has an error of only 168 pixels so the
MAE is tiny, however the effects are eye-catching and make the edges of the ob-
jects look “furry”. A median filter makes little impression reducing the error to 153
pixels. A further iteration has little effect (150-pixel error). The optimum filter in a
5 × 5 window, however, reduces the error to just 34 pixels. It would be difficult to
“guess” a filter that would perform close to the optimum for this type of example.