Page 147 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

124 Chapter 4. Basic Motion Estimation Techniques

resulting bits were included in the bit-rate calculations. All quoted results refer
to the luma components of sequences. Figure 4.15 compares the performance
of the three algorithms when applied to the three test sequences.
In general, both interframe coding algorithms (FDIFF and BMA-H) out-
perform the intraframe coding algorithm (INTRA). Thus, even at very low
bit rates, high frame skips, or low-motion sequences, the temporal correlation
between video frames is still high enough to justify interframe coding.
Comparing the two interframe coding algorithms, it is immediately evident
that the BMA-H algorithm outperforms the FDIFF algorithm at all bit rates
and for all sequences. Note, however, that at extremely low bit rates, and in
particular for the low-motion AKIYO sequence, the e!ciency of the BMA-H
algorithm starts to drop and approaches that of the simpler FDIFF algorithm.
But even with this drop in performance, the use of BMA-H is still justi able.
For example, with AKIYO and at a bit rate as low as 3 kbits=s, the BMA-H
algorithm still outperforms the FDIFF algorithm by about 1 dB.

4.9 Discussion

Motion estimation is an important process in a wide range of applications.
Di erent applications have di erent requirements and may, therefore, employ
di erent motion estimation techniques.
In video coding, the determination of the true motion is not the intrinsic
goal. The aim is rather to simultaneously minimize the bit rate corresponding
both to the motion parameters (motion bits) and to the prediction error signal
(DFD bits). This is not an easy task, since the minimization of one quantity
usually leads to maximizing the other. Thus, a suitable tradeo is usually
sought. In this chapter, four motion estimation methods were compared. The
four methods are the di erential, pel-recursive, phase-correlation, and block-
matching motion estimation methods. It was found that block-matching motion
estimation provides the best tradeo . It uses a block-based approach to reduce
the motion overhead while still maintaining a very good prediction quality
(and consequently a small number of DFD bits). This explains the popularity
of this approach and its inclusion in video coding standards.
The chapter also investigated the e!ciency of motion estimation at very low
bit rates. It was found that the prediction quality of motion estimation starts
to drop at very low bit rates, in particular, for low-motion sequences, and
approaches that of simpler techniques, like frame di erencing and intraframe
coding. Despite this drop in prediction quality, it was found that the use of
motion estimation is still justi able at those bit rates.