Page 145 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

122 Chapter 4. Basic Motion Estimation Techniques

4.8 E5ciency of Block Matching at Very Low Bit
Rates

The incorporation of motion estimation and compensation into a video codec
involves extra computational complexity. This extra complexity must, there-
fore, be justi ed on the basis of an enhanced coding e!ciency. This is very
important for very-low-bit-rate applications and, in particular, for applications
like mobile video communication, where battery time and processing power
are scarce resources.
Very low bit rates are usually associated with high frame skips. As the
frame skip increases, the temporal correlation between consecutive frames de-
creases. This will obviously decrease the e!ciency of motion estimation, as
can be seen in Figure 4.13. This poses a very important question: Is the use
of motion estimation at such bit rates justi able? Or put in another way, Is the
use of less complex coding methods, like frame di erencing and intraframe
coding, su!cient at those bit rates?
This study investigates the e!ciency of block-matching motion estimation
at very low bit rates. Three algorithms were implemented:
BMA-H This is a half-pel full-search BMA with 16 × 16 blocks, ± 15 pels
maximum displacement, restricted motion vectors, and SAD as the match-
ing criterion. Half-pel accuracy is achieved using a re nement stage around
the full-search full-pel motion vectors. Bilinear interpolation is used to
obtain intensity values at subpel locations of the reference frame.
FDIFF This is a frame di erencing algorithm. This means that no motion
estimation is performed and the motion vectors are always assumed to
be (0; 0). Note that this algorithm has no motion overhead and the total
frame bits are equal to the DFD bits.
INTRA This is a DCT-based intraframe coding algorithm.
In each algorithm, motion was estimated and compensated using recon-
structed reference frames. Motion vectors were coded using the median pre-
dictor and the VLC table of the H.263 standard. Both, the DFD signal (in
the case of BMA-H and FDIFF) and the frame signal (in the case of
INTRA) were transform encoded according to the H.263 standard. To simulate
a very-low-bit-rate environment, the frame skip was set to 4 (this corresponds
to 7:5 frames=s for AKIYO and TABLE TENNIS and to 6:25 frames=s for FOREMAN).
To generate a range of bit rates, the quantization parameter QP was varied
over the range 5–30 in steps of 5. This means that each algorithm was used to
encode a given sequence six times. Each time, QP was held constant over the
whole sequence (i.e., no rate control was used). The rst frame of a sequence
was always INTRA coded, regardless of the encoding algorithm, and the