motion search precision

[fiercekid84]

for SVCD should i use highest quality for CBR & VBR or motion estimate search and also for VBR long substained format ( out of standard ) which will it be best???



and how about VCD use motion estimate search or highest quality ???????

[Cobra]

Fiercekid84,

I believe the general opinion is that highest quality motion search is a waste of time for the tiny amount of extra quality it gives. I personally use high quality for a balance between quality and time.

It does not matter what standard you are using, motion search is motion search. If it is not done to a high standard then it will produce substandard video.

I am not sure what actual difference it makes - can someone else shed some light here?

CobraDMX

[AdamRA4]

I have always been in question what 'motion search precision' actually does. Can you enlighten?

[adam]

To understand this setting you have to have a basic understanding of mpeg encoding. Its all about removing reduncy. If you look at any given sequence of frames taken from a movie you will see than many of these frames look almost identical, or at least share mostly the same qualities, ie: still background but moving person. So the idea is, why store all that redundant information in each frame? Instead, mpeg has I frames which represent 1 unique picture. Then there are a series of frames which only contain the information which differs from the I frame. So say we have a sequence of frames with a man standing in front of a stationary background. Nothing changes from 1 frame to the next except that his hand moves. So the I frame would just be a picture of the entire scene. The next frame would just contain the little bit of hand that differs, and so on. During decoding, each of these partial frames draws the information it needs from I frame and reconstructs the entire picture. The result is movement.

So you can see that the majority the compression gained in mpeg encoding is just recognizing movement and throwing out everything else that doesn't contribute to it. So this is what the motion search precision option does. It determines how strenuously, and thus how long, the encoder spends comparing 1 frame to the rest to determine, essentially how much redundancy there is.

This setting is clearly important. The better the motion recognition, the better compression you get at any bitrate level, or in other words the more useful data you are able to keep. This setting is basically nothing more than a quality vs encoding time adjustment.

[SaSi]

And to add to what Adam stated, the lower the encoding bitrate, the more important motion search precision becomes.

At 8000kbps encoding rate, motion search becomes much less important because the encoder has enough bitrate budget to allocate to each frame without loosing quality. So people fitting 1 hour's worth of video to a DVD will say motion search is not important.

At 1150kbps, motion search becomes (I dare say) critical, since if picture redundancy is not remove, it has to be encoded at the expence of overall quality.

So, the lower the bitrate you need to encode to, the higher the search precision you must select for acceptable quality.

Additionally, the quantization matrix is equaly important depending on bitrate. At high bitrates again it's importance is lower, but if you aim for encoding at 2000kbps or lower average, then the quantization matrix must be chosen carefully.

CCE has special QM values for ultra low bitrates (as it calles them). Several threads have discussion on this subject.

[AdamRA4]

Thanks for breaking it down.

[cooljoe666]

Whatever....I have tried every setting on the same source file and now I leave it on SLOW-HIGH QUALITY with every conversion.

[iThinkYouBrokeIt]

Hmm...

i found reading this post quite helpful.

I've always wondered what motion search precision was, and exactly what difference it makes.

Perhaps I should use a higher motion search precision setting when encoding MPEG-2 352x480 @ 950kbps...

[SaSi]

For really low bitrate encoding (and 950kbps is real low for 352 x 480), you should also consider modifying the Quantization Matrix coefficients as well.

CCE has two additional settings for very low and ultra low bitrate.

I have found that using these with very very low bitrates causes encoding to soften the image a bit rather than cause pixelation in high motion sequences. Between pixelation and smoothness, I prefer the second.

You can use these coefficients in any other encoder that allows you to enter them (like Tmpgenc or Mainconcept).

I feel bad not commenting about Canopus Procoder that is highly considered by those that use it but not as often spoken about as the other two. (Any procoder users to comment if QM coeffs can be modified there?)

Additionally, with lower bitrate you could perhaps decrease precision from 10 or 9 bits down to 8. Higher precision calls for a higher bitrate to actually make any difference (this is what I gather rather than understand about it). Also, at least with Tmpgenc, I've noticed a slight increase in performance with 8 bits compared with 10 bits - although nothing like the speed increase from High quality to low quality motion search mode.