CPP100
SECTION 1
CPP1OPS
25/10/06
www.snellwilcox.com
Version 1 Issue 6
1.5
Pre Processor Overview
Recursive Filter
Recursive filters reduce noise by temporally
averaging successive pictures. Utilising delays of
exactly one picture or frame, noise can be reduced in
stationary areas without loss of spatial (horizontal and
vertical) resolution. Although temporal recursive
filters offer considerable levels of noise reduction,
sophisticated control logic is required to ensure that
picture detail is preserved at higher noise settings.
In particular, analysis of the noise floor level is
necessary to set movement thresholds at levels that
are just above the noise floor. At optimum settings
this
allows
maximum
noise
reduction
and
simultaneously maximum sensitivity to movement.
Auto Threshold Bias
In auto threshold mode the noise detection algorithm
may be given a subjective bias to give more or less
noise reduction. Modification of the bias should not
be necessary under normal circumstances.
Y And C Recursive levels
These settings change the amount of noise reduction
for luminance (Y) and chrominance (C) by limiting the
maximum level of noise reduction. The actual level
of noise setting is dynamically adjusted on a pixel-by-
pixel basis with regard to the noise setting for the
same pixel in the previous frame. Other factors such
as movement contribute to the current pixel setting.
This mechanism ensures that the optimum level of
noise reduction is applied to each pixel.
Threshold
This sets the threshold for the motion detector. The
lowest level of 0 gives the greatest sensitivity to
motion, but allows more noise to break through, while
15 gives the greatest noise reduction but can lead to
excessive filtering of low-level textures. When this is
set to auto the threshold is dynamically set to an
appropriate value for the current input noise level.
Semi Transversal
The semi-transversal filter is a uniquely patented
design that operates in conjunction with the recursive
filter to increase its effectiveness. Quite unlike
traditional transversal filters it operates by selecting
the most appropriate outputs from a chain of picture
stores at the output of the recursive filter.
An algorithm is used to determine which of the stores
contains the highest level of noise-reduced picture.
The overall effect is to increase the amount of noise
reduction in a typical picture. For example, moving
objects cause the recursive filter to turn off at the
edge of the moving object. This leads to a
recurrence of noise that takes a number of frames to
reduce to the defined user level. The semi-
transversal filter is able to monitor the recurrence of
noise and delay the output of the recursive filter up to
a maximum of three frames. Operating on a pixel-by-
pixel basis, the overall level of noise reduction in a
typical picture is maintained at a more uniform level
and is less dependent on movement.
As the semi-transversal filter complements the
recursive filter, it cannot be utilised without the
recursive filter. Effective at all recursive filter settings
its operation can be seen as a reduction in the level
of revealed noise trail following moving objects.
The semi-transversal filter operates in a fully
automatic mode - there are no user adjustments
required.
Median Filter
Median filters can be effective at removing impulse
noise. They operate by rank filtering pixels from an
odd number of aperture points yielding the median
value. The aperture set may utilise the surrounding
pixels from the same field or more usually some
combination of pixels from current and adjacent fields
or frames.
When a pixel is judged to be in error it is replaced by
the median value of the aperture set. Pixels judged
not to be in error remain unaltered. The algorithm is
therefore quite specific about the areas of the picture
which are filtered.
An algorithm utilises both spatial and temporal
gradient information to determine if the suspect pixel
has impulse noise characteristics.
