Digital Signal Processing Reference
In-Depth Information
processing. This allows for compression in the higher-quality
chroma-resampling format.
15.1.1.2 MPEG-2 Interlaced Video Support
As discussed in earlier chapters, interlaced video is a legacy of
television broadcast systems, used to raise the frame-update rate
to avoid flicker while not increasing the actual data rate. Inter-
laced video is challenging for compression, because the alter-
nating rows are from different time instants and can have
different motion vectors, so that the optimum predictive match
may be from different frames. MPEG-2 adds techniques to cope
with interlaced video, some of which are listed below.
15.1.1.3 MPEG-2 Frame and Field DCT
MPEG-1 uses a frame-type DCT; that is the DCT is performed
on 8
8 arrays of pixels in the same frame. This would be sub-
optimal when using interlaced video, as differences between the
interlaced pixel rows in the frame could introduce substantial
high-frequency content into the DCT coefficients, leading to poor
compression.
MPEG-2 has an option to use field DCT. When using field DCT,
each of the four 8
16 macrocell array has the
DCT performed independently, resulting in four DCT coefficient
arrays of 8
8 blocks in the 16
16 macrocell is
divided into two 8 wide by 16 tall pixel arrays. An 8
8 each. For interlaced video, the 16
8 DCT
transform is performed on the odd rows (1, 3, 5, 7) of each 8
8
pixel array. This results in the top array of 8
8 DCT coefficients.
A second 8
8 DCT transform is then performed on the even
rows (2, 4, 6, 8) of each 8
8 pixel array, and is used to compute
the bottom array of 8
8 DCT coefficients.
In this way, the interlaced sets of rows, which correspond to
the two different instants in time of the video frame, are trans-
formed and quantized separately.
15.1.1.4 MPEG-2 frame and field motion prediction and estimation
Progressive video uses the 16
16 macroblock to perform
prediction and motion estimation. When using the field DCT it
also makes sense to perform the motion estimation using each
16
8-sized pixel array corresponding to each field of the inter-
laced video macroblock.
15.1.1.5 MPEG-2 Quantization
Additional bits can be used for DCT coefficients, beyond the
eight bits used in MPEG-1. Up to 10 bits can be used.
