1999 May 03
13
Philips Semiconductors
Product specification
Picture Improved Combined Network
(PICNIC)
SAA4978H
7.2.11
N
OISE REDUCTION
The noise reduction part consists of clamp noise reduction
and spatial noise reduction for low frequency noise. Within
this ensemble a two dimensional band split is used,
enabling also the functions of 2D low passing, adding the
multi Picture-In-Picture (multi PIP) function and 2D
peaking.
The clamp noise reduction is realized with an adaptive
temporal recursive filter. This filter will correct the DC level
of each line when it is varying from field-to-field in the
segments with the least likely movement. This clamp noise
filtering is intended to correct for clamp errors in a
complete chain, which cannot be removed with traditional
clamping on the back porch of the video. Clamp noise is
only reduced for luminance.
The spatial noise reduction is targeted for reduction of the
mid frequency noise spectrum, where adaptive filtering
combines pixels around the centre pixel and pixels from
the lines above in a recursive way. This spatial noise
reduction is only realized for luminance.
The 2D low-pass filter is a [1; 2; 1] filter in both the
horizontal and vertical direction. 2D high-pass is realized
by taking the centre tap and subtracting the 2D low-pass
output from it. Also added in the 2D high-pass is the
vertical low-passed data, which is subtracted from the
centre tap and multiplied by a user selectable gain
(0 to
7
8
). The 2D high-pass data is multiplied by a user
selectable gain of 0 and
2
4
to
8
4
and cored before adding
it to the 2D low-pass branch for the 2D peaking function.
The HF signal bypasses both the LF temporal and the
spatial noise reduction, therefore sharpness in the high
frequencies is not reduced by the noise reduction parts.
The factor 0 on the HF signal yields a pure 2D low-passed
signal at the output. Multi PIP with pure subsampling of this
signal yields a much better result than without the low-pass
operation.
7.2.12
H
ISTOGRAM
Histogram modification consists of acquiring the histogram
of the luminance levels and correcting the luminance
transfer curve in order to provide more perceptual contrast
in the picture.
For economy, a subsampling is realized on the video with
a factor of 4 before the histogram is produced. From
line-to-line, a two pixel offset is used on the subsample
pattern.
The histogram acquisition uses 32 baskets on the grey
scale from (ultra) black to (ultra) white. Pixels that are
found around the centre of a basket increase a counter for
that basket with the value 8, pixels that come around the
edge between two baskets increase the counters in both
baskets, such as 3 in the left one and 5 in the right one.
By this method, the quantization distortion is overcome
from having a discrete set of baskets.
Between acquisition of the histogram and correction of the
transfer curves, the microcontroller included in the
SAA4978H processes the counter values from the
32 baskets. The outcome of the microcontrollers algorithm
defines a differential transfer curve for the luminance. This
means that only differences from a 1 : 1 transfer curve are
coded. This is done in 32 LUT points, with a linear
interpolation for all input values in between the LUT points.
When changes are made to the luminance level of pixels,
the saturation has to be restored by using the same
relative gain for the U and V channels.
The histogram data also provides the information of the
minimum and maximum levels of Y, U and V, by which the
microcontroller can affect an AGC gain before the video
analog-to-digital conversion.
Another main part of the histogram is the display-bars
block. This block can insert up to 32 horizontal bars in the
YUV data path. Size, spacing, luminance, colour and
length are fully programmable. This can be used to
construct a visual display of the histogram or transfer
curve.
7.2.13
S
UBTITLE DETECTION
Subtitle detection searches in a large area of the video
field for patterns that are characteristic for subtitles.
The expectation is to encounter in a video line a
considerable number of crossings through both a dark
grey and a light grey threshold and in its vicinity also
crossings in the other direction. This part is realized with
valid crossing (event) counting on each line in the target
area. This event value is stored for 128 lines in the subtitle
RAM, which is located at the top of the auxiliary RAM.
The subtitle logic has higher priority to access the subtitle
RAM than the microcontroller.
The internal microcontroller can filter out this data. In a
number of adjacent lines, there must be a similar high
count value for the number of events. If this condition holds
then the detection of subtitles on that vertical position is
more definite.