Image Processing Reference
In-Depth Information
The main function of the median operator is to remove noise (especially salt and pepper
noise) whilst retaining the edges of features in an image. You can't see that here, there is
little image noise. So let's add some in:
noisy_p:=addcondiments(eye,0.1)
If you make the value supplied as an argument to
addcondiments
smaller, you'll get
less
noise, larger values (0.3 say) result in greater noise contamination.
normalise (noisy_p)
10/10 for the label of this image! Now we have introduced light (salt) and dark (pepper)
points into the image. This type of noise is quite common in satellite image transmission
decoding errors.
So let's see what our median operator can do with this image, in comparison with direct
and Gaussian averaging:
Median Mean = Direct Averaging Gaussian Averaging
nmed:=med(noisy_p) nmean:=ave(noisy_p,3) gmean:=tm_conv(noisy_p,
Gaussian_template(3,0.8))
normalise(nmed)
normalise(nmean)
normalise(gmean)
(a)
(b)
(c)
The median operator clearly has a better ability to remove this type of noise. This is
because it is removed when it is placed at either end of the rank sorted list. However, in
