Digital Signal Processing Reference
In-Depth Information
residuals, or residual artifacts. These artifacts can be determined
by differences in the four 8
8 DCTs in the macroblock and then
coded and compressed as part of the frame data. Of course, if the
residual is too large, coding the residual might require more data
compared to just compressing the image data without any
reference to the previous frame.
To begin MPEG compression, the source video is first con-
verted to 4:2:0 format, so the chrominance data frame is one
fourth the number of pixels
½ vertical and ½ horizontal reso-
lution. The video must be in progressive mode: each frame is
composed of pixels from the same time instant
e
(i.e. not
interlaced).
At this point, a bit of terminology and hierarchy used in video
compression needs to be introduced:
A pixel block refers to an 8
8 array in a frame.
A macrocell is 4 blocks, making a 16
16 array in a frame.
A slice is a sequence of adjacent macrocells in a frame. If data
is corrupted, the decoding can typically begin again at the next
slice boundary.
A group of pictures (GOP) is from one to several frames. The
significance of the GOP is that it is self-contained for compres-
sion purposes. No frame within one GOP uses data from
a frame in another GOP for compression or decompression.
Therefore, each GOP must begin with an I-frame (defined
below).
A video is made up of a sequence of GOPs.
Most video compression algorithms have three types of
frames:
I-frames
they are
compressed using only information in the current frame.
A GOP always begins with an I-frame, and no previous frame
information is required to compress or decompress an
I-frame.
P-frames
e
These are intra-coded frames
e
These are predicted frames. P-frames are
compressed using image data from an I- or P-frame (that
may not be the immediately preceding frame) when compared
to the current P-frame. Restoring or decompressing the frame
requires compressed data from a previous I- or P-frame, and
residual data and motion estimation corresponding to the
current P-frame are used. The video compression term for
this is inter-coded, meaning the coding uses information
across multiple video frames.
B-frames
e
These are bi-directional frames. B-frames are
compressed using image data from preceding and successive
I- or P-frames. This is compared to the current B data to
e
