[Mp4-tech] [H.264]Doubt regarding search range

Tourapis, Alexis alexis.tourapis dolby.com
Thu Dec 28 13:12:17 ESTEDT 2006


Dear Hari,
There are many factors that can determine search range. These include
motion within the sequence, resolution, distance between references,
complexity/power requirements of the device you are implementing, motion
estimation (ME) and mode decision algorithms you may be using etc.
Different ME schemes for example may be invariant and even indifferent
of the search range used as well (this applies mainly to software
implementations) primarily in terms of implementation complexity. Note
that for the JM, the search window is in fact a moving window which has
its center adjusted according to the motion vector predictor used for
coding. That essentially considerably increases the "dynamic" range of
motion that one may find when encoding a sequence. Btw, the number 16 is
only an example and it not really a suggested value. Old JVT common
conditions applicable to QCIF and CIF resolutions considered a search
range of +-32, but that increased based on resolution later on.
Finally, there is an old paper by Gonzales et al. discussing some issues
relating to motion search range. Even though this was primarily targeted
to MPEG-2, it still has some relevancy with MPEG-4 AVC/H.264.
Mean absolute Error or Mean absolute Difference (MAD) are just
normalized versions of the  SAE/SAD. That is you are basically dividing
the value of SAD with the number of pixels that exist within a block. As
you can understand there is no reason to perform such a normalization
since such would be applied equally among all blocks that you will be
searching (this is still the case even for AVC where you have multiple
blocks since in general mode decision is performed at the
macroblock/subblock level with the same number of pixels). MSE or the
equivalent SSE are more complex to compute than SAD due to the
multiplication required (this can be compensated using table look up
implementations) but also memory. MSE essentially can be seen as the L2
norm of the residual after motion estimation, while MSE can be seen as
the L1 norm. Note that unfortunately the MSE does not conform to the
Human Visual System (HVS) and therefore there is little benefit in using
this while it can also be at times sensitive to noise. On the other
hand, some hardware implementations actually provide instructions that
can perform SAD operations in a single cycle (i.e., SIMD on generic
CPUs). For your reference, the upcoming version (12.0) of the JM will
allow you to also use MSE for motion estimation.
I am not very clear on what you define as spatial and temporal
information. If you are talking about texture/edge characteristics vs.
sad values then such may be completely uncorrelated. If however you use
older "spatial" information (i.e. you have performed spatial analysis on
earlier frames) such could help you characterize the spatial information
of future frames.
Best regards
Alexis
________________________________
From: mp4-tech-bounces lists.mpegif.org
[mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of hari prasath
Sent: Thursday, December 28, 2023 6:23 AM
To: mp4-tech lists.mpegif.org
Subject: [Mp4-tech] [H.264]Doubt regarding search range
Dear Experts,
I have a three doubts..
1) In the JM reference Encoder,search range of 16 is used.Is there any
specific reason for it.What are the factors that decide the search
range in motion estimation. 
2)What is the difference between the energy measures Mean Square
Error(MSE),Mean Absolute Error(MAE) and Sum of Absolute differences(SAD
or SAE)?In the reference encoder,SAD is used.Why not MSE or MAE?
3)This is a general doubt.Is it possible to get the spatial information
in a frame from the temporal information in a video sequence i.e are the
spatial information and temporal information orthogonal?
Could anyone help me...
Thanks in advance... 
-----------------------------------------
This message (including any attachments) may contain confidential
information intended for a specific individual and purpose.  If you
are not the intended recipient, delete this message.  If you are
not the intended recipient, disclosing, copying, distributing, or
taking any action based on this message is strictly prohibited.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20061228/5651a4bb/attachment.html


More information about the Mp4-tech mailing list