Image Processing Reference
In-Depth Information
This gives the famous
bell-shaped
function shown here for a 19× 19 window with a
standard deviation of 4. Try changing the standard deviation from 4 to, say, 2 and 8 so you
can see its effect on the width.
Gaussian_template (19, 4)
So let's apply it:
Gaussian_smoothed:=tm_conv(eye,Gaussian_template
(3,0.8))
And the result is:
normalise (Gaussian_smoothed)
This can keep much more detail concerning image features; note here its ability to retain
detail in the eye region which was lost in the earlier direct averaging. Again, it can be
implemented in the frequency domain, as can any template convolution process.
The mean and Gaussian operators are actually
statistical
operators since they provide
estimates of the mean. There are
other
statistics; let's go for a
median
operator. This gives
the midpoint of a sorted list. The list is derived from the pixel values within a specified
area. We need to provide the
sort
function with a vector, so for a
3
×
3 neighbourhood
centred on a point with co-ordinates
x
,
y
, we get
y:=3 x:=3
