|
The normalisation
filter is an non-linear filter useful to detect dark narrow bands with a
background of varying intensity. In such cases it can be difficult to set a
threshold, especially if you search for threshold passing, but also if you
search for intensity jumps. In both cases the varying background is a
problem. The normalisation filter flattens the background, thus the dark
bands turns out with their relative strength compared to the background. If
they have similar relative strength compared with the background, a common
detection threshold will work.
The general
background intensity in a neighbourhood of all points on the intensity
profile is calculated. Then the point’s intensity is divided by the
background intensity and multiplied with 100. You get an intensity profile
consisting of percent values close to 100. Only where dark bands it will be
below 100. The bands can know be measured as threshold passing if setting a
threshold less than 100, e.g. 80.
To parameters
manages the normalisation. The normalisation width decides the width
of the point neighbourhood. The number sets how many points to each side you
calculate. A normalization width of 5 means for example that the intensity
is measured for a total of 11 points. All points in the neighbourhood are
sorted by increasing intensity. The normalisation selector decides
which of the sorted points to use as background value. If we use the example
with normalisation width set to 5, the normalisation selector must be in the
–5..5 area. –5 picks the lowest of the sorted numbers, 0 picks the
median and 5 picks the largest number. Other values pick a number in
between. If you have a large normalisation width (most of the neighbourhood
points having a background intensity), 0 will e a robust and good normalisation selector. If you have a narrower width and only a few of the
points having any background intensity, you must use a higher selector value
since you want to pick one of the higher values. Avoid however the maximum
value since it often is noisy.
The normalisation
filter can likewise be used for light bands on varying background. For
example to find weaknesses in a transparent material with varying background
light.
Parameters guiding filtering
|
Filter
parameters
|
Description
|
|
Smooth base
|
Sets the
filter width, or how large the smoothing neighbourhood is. Increasing
this value does not considerably increase the time of analyses.
|
|
Smooth counter
|
Sets the
number of smoothing iterations. In praxis the filter turns wider. The
time of the analyses linearly increases with this parameter.
|
|
Differentiation
|
Sets the
differentiation order.
The values
are
0.
No differentiation
1. First derived (gradient)
2. Second derived (curve)
The time of
the analyses linearly increases with the order.
|
|
Normalisation
width
|
Sets the
width of the neighbourhood calculating the background intensity. The
time of the analyses linearly increases with the width.
|
|
Normalisation
selector
|
Selects the
background intensity from a sorted list of neighbourhood values
decided by the normalisation width k. Valid values are then
–k..k.
This parameter does not affect the time of analyses.
|
|
