![background image](http://html1.mh-extra.com/html/snell-and-wilcox/prefix-cpp100/prefix-cpp100_operators-manual_1309061083.webp)
CPP100
SECTION 9
CPP1OPS
25/10/06
www.snellwilcox.com
Version 1 Issue 6
9.1
Appendix 1-Using the Filters
Recursive Filtering
Noise can be reduced in video signals without impairing spatial (horizontal and vertical) resolution by using the
fact that in stationary pictures the only difference between successive frames is caused by the noise itself.
Thus temporal averaging of successive video frames will produce a degree of noise reduction of the picture
since the noise contribution is random with zero mean. The overall effect of this is a temporal low pass filter.
There are two ways of implementing such a low pass filter. Firstly the signal may be passed through a series of
delay elements, each of length one picture period, and the output signals summed together as shown below:
This type of noise reducer is typically known as a transversal noise reducer. The level of noise reduction that
can be obtained from this type of arrangement is directly determined by the number of temporal or picture
contributions. Unfortunately, a large number of picture contributions are required to achieve a useful level of
noise reduction. Another consequence of this arrangement is that it introduces a substantial processing delay.
An alternative filter arrangement is shown below:
The structure of this filter is basically a two input cross-fader between the video input and previous output
delayed by one frame. The cross-fade value, K is determined by factors such as the degree of noise reduction
required and the level of picture difference detected between current input and frame delayed output. Hence if
the K value is very small then the contribution to the output will predominantly be from the previous output and
not from the current input. On a static scene following a shot change the picture output will be formed entirely
from the input (K=1) since there will be a large picture difference between scenes. On the next frame however,
K will be reduced to the optimum value of 1/2 and some noise reduction will be accrued from the combination
of current input and previous output. On the following frame the value of K will be further adjusted to the
optimum value of 1/3 and so forth for additional frames until K reaches the lowest corresponding to the
ultimate level of noise reduction required.