Compression - 3DTV: Processing and Transmission of 3D Video Signals - page 69

Prediction
error

Macroblock of
input image

Quantized
coefficients

+

Transform Quantization

Entropy coding

Inverse
transform

+

Deblocking
filter

Intra-frame
prediction

Motion
data

Motion
compensated
prediction

Memory

Intra/Inter

Motion
estimation

Figure 3.2 High-level encoder block diagram of AVC [3]

I-slices are encoded by intra-prediction, whereas P and B slices are encoded
by inter-prediction. P slices can use references from a list of preceding
reconstructed frames, whereas B slices can use references from two lists of
reconstructed pictures: from preceding frames and from succeeding frames
in time. The SI and SP slices are especially used for efficient bit-rate switching.
More detail on SI and SP frames can be found in [2]. A single frame may
contain a mixture of these slices.
Intra-coding is a scheme that does not exploit temporal redundancies
existing among successive frames of a video sequence. The aim of intra-
coding is to provide error resilience and create temporal random access points
in a video sequence, more than to provide coding efficiency. Different from
the previous coding standards, AVC applies a spatial estimation method
for intra-coding. Spatial estimation is performed within the same frame
using already encoded macroblocks. Intra-coding is especially efficient in
compressing texture-wise smooth areas inside a frame.
Inter-coding aims to remove the temporal redundancies inside the video
sequence via block motion estimation, as previously explained. There are
two kinds of inter-frame coding in AVC: P (predictive) and B (bi-predictive)
coding. In both techniques, the reconstructed samples from previously
encoded frames are used. In previous standards, only the frame, which is
the nearest frame to the currently encoded frame in time, could be used for
P prediction. However, in AVC, multiple previously encoded frames can be
used as the reference for P prediction. In B prediction, the encoded frames,
which are encoded previously but succeeding the currently encoded frame in
display time, can be used as references for prediction. However, this brings
the cost of larger coding delays and increased memory usage, while yielding
much higher compression performance.