Texas Instruments TMS320C64x DSP Benutzerhandbuch

4-22

Chrominance resampling computes chrominance values at sample points
corresponding to output luminance samples based on the input interspersed
chrominance samples. This filter performs the conversion between inter-
spersed YCbCr 4:2:2 format and co-sited YCbCr 4:2:2 format. The vertical
portion of the conversion from YCbCr 4:2:0 to interspersed YCbCr 4:2:2 must
be performed in software.

The chrominance resampling filters calculate the implied value of Cb and Cr
co-sited with luminance sample points based upon nearby interspersed Cb
and Cr samples. The resulting values are clamped to between 01h and FEh
before being output. Chrominance resampling is shown in Figure 4–19.

YCbCr 4:2:2 interspersed

source pixels

YCbCr 4:2:2 co–sited

output results

Luma (Y)

sample

Cb’f = (–3Cbab + 33Cbcd + 101Cbef – 3Cbgh ) / 128
Cr’f = (–3Crab + 33Crcd + 101Cref – 3Crgh ) / 128

Chroma (Cb/Cr)
samples

–

a

b

c

d

e

f

g

h

i

j

k

l

–

The 2

×

-scaling mode is used to double the horizontal resolution of output

luminance and chrominance data. This allows processed CIF resolution images
to be output at full size. Vertical scaling must be performed in software. Scaling
for co-sited source is shown in Figure 4–20 and scaling for interspersed
source is shown in Figure 4–21.

For a co-sited source, the source luminance pixels are output unchanged for
every even pixel (a, b, c, etc., in Figure 4–20). Odd luminance pixels (a’, b’, c’,
etc.) are generated from neighboring source (even) pixels using a four tap fil-
ter. The chrominance source pixels are output unchanged for every other even
pixel (a, c, e, etc.). Other even output pixel (b, d, f, etc.) chrominance values
are generated from neighboring source chrominance pixels using a four tap
filter.

For an interspersed source, the luminance is output identically to the co-sited
case. Chrominance output is generated using a four tap filter with one of two
different coefficient sets depending on which source chrominance pixel the
output pixel is closest.

Note that because input scaling is limited to 2x, full BT.656 width output is not
achieved from CIF source images. The horizontal location of the reduced
image can be adjusted using HOFFSET.