
TM1100 Preliminary Data Book
Philips Semiconductors
13-10
PRELIMINARY INFORMATION
File: icp.fm5, modified 7/26/99
The ICP does not have full chroma keying. Full chroma
keying has adjustable threshold values for the pixel com-
ponents. Adjustable thresholds allow the user to auto-
matically select an overlay sub-image from a larger over-
lay background, such as selecting an image of an actor
against a bright blue background while inhibiting the blue
background.
13.5.6
Dithering
Short output codes, such as RGB 8, have few bits for out-
put value determination. RGB 8R has (2,3,3) bits for
(R,G,B). The result is a coarse, patchy image if nothing
is done to correct for the limited resolution. Dithering sig-
nificantly improves the effective resolution of these imag-
es. RGB 8 images with dithering look nearly as good as
RGB 16, for example.
Dithering works by adding a random dithering value to
the pixel before it is truncated by the output formatter.
The dither is added to the portion which will be truncated.
The carry from this add will occasionally propagate into
the most significant portion of the pixel before truncation.
The carry from the add thus “dithers” the displayed value.
random dither value is added to the original data before
truncation. The dither value should have a range of from
approximately 0 to 1 LSB of the truncated value. The
dither value should be symmetrical about 1/2 the LSB of
the quantizing error of the truncation. In the example
shown, the dither signal has values of (1/8, 3/8, 5/8, 7/8).
This set of values has a range of approximately 0 to 1
LSB, and it is symmetrical about 1/2 LSB.
In this example, the input signal has a value of 2.83.
Without dithering, this value would be truncated to an
output value of 2 in all cases. Averaging the un-dithered
signal over four pixels still gives you a value of 2. By add-
ing the dither signal, the output value is 2 or 3 depending
on the value of the added dither signal. Averaging over
four pixels, the average output value is 2.75, much closer
to the input value than without the dither signal. The dith-
er signal has significantly reduced the error when aver-
aged over four pixels.
Two types of dithering are combined in the ICP: quad
pixel and full image dithering. Quad pixel dithering, also
known as ordered dithering, adds one of four dithering
values to each pixel. The four dithering values corre-
spond to four-pixel quads in the output image. The pixels
in each quad have fixed positions in the input image, so
the dither values are chosen on the bases of odd or even
line number and odd or even pixel number in the line.
The dither values of (0/4, 3/4, 2/4, 1/4) are added by line
and pixel number: (even line & even pixel, even line &
odd pixel, odd line & even pixel, odd line & odd pixel).
This gives a four value ordered function for four adjacent
pixels in the image. The (0,3,2,1) pattern is chosen spe-
cifically to prevent pairs of high or low pixel values from
clustering. Spatial dithering provides a significant im-
provement in effective resolution.
Full image dithering adds a random number to every pix-
el of the image. The result is that the intensity and color
accuracy increases as the size of the sample is enlarged.
The random number has a long bit length to prevent re-
peating patterns in the image. The random number can
be static or dynamic. In the static case, the random num-
ber generator starts with a fixed seed at the start of the
image. The random number spatial pattern is fixed for
the image even though the image data may change from
frame to frame. In the dynamic case, the random number
generator runs continuously, and the dithering pattern
changes from frame to frame.
The ICP adds full image dithering to the quad pixel dith-
ering to provide the final dithering signal for each pixel.
The quad pixel dither provides the two most significant
bits of the dither signal, and the full image dither provides
the least significant 4-bits of the dither signal. The com-
bined dither signal is 6 bits.
From 1 to 6 bits of dither signal are used, depending on
the output format. If fewer than 6 bits are needed, only
the most significant bits (MSBs) of the dither signal are
used. For example in the RGB8R output format, the R
output value is 3 bits in size. The output uses the 3 MSBs
of the R input value and truncates the 5 LSBs. The dither
unit adds 5 bits of dither signal (the 5 MSBs) to the 5
0
1
2
3
2.830
Dither = 0
Output = 2
0
1
2
3
2.955
Dither = 1/8
Output = 2
0
1
2
3
3.205
Dither = 3/8
Output = 3
0
1
2
3
3.455
Dither = 5/8
Output = 3
0
1
2
3
3.705
Dither = 7/8
Output = 3
No Dithering:
Output = 2.0
1/4 LSB Dithering
Output = (2+3+3+3)/4 = 11/4 = 2.750
Error = +0.830
No Dithering
1/4 LSB Dithering
Error =(2.830 - 2.750) = +0.080
Figure 13-10. Dithering