Encoder Precision Vs Bit Rate

[ArmandLiberi]

While waiting for tech support on other issues, I decided to try to learn a little about the Recode capabilities. I started by recoding the same video 11 times to an FLV file with the encoder precision varied from 0.0 to 1.0 in 0.1 steps. I was a little surprised to find that the output bit rate was not proportional to the encoder precision. Thinking that this is not correct, I wrote the following to tech support:

"After noticing unexpected results in Nero Recode with a disc structure, I made a single frame video with Nero Video and recoded it to 11 different .FLV files, varying encoder precision. I have attached the original file, the log files, and the single frame output files. In addition, see the output of the Media Info program (MediaInfoOriginal.txt). I have condensed the MediaInfo info into an Excel file, showing only differences (20120413_MediaInfoDifferences.xls). Furthermore, I have reduced that to a single Word file (EncoderPrecisionVsBitRate.doc) showing that output bit rate is not proportional to the encoder precision setting as an example of what I consider "unexpected." (There is also a MediaInfo file for the original larger output files, MediaInfoOnOriginalFiles.txt.

"Is this correct?"

The attachments are attached to this posting also for the curious. I have posted this here for two reasons:

1. If I am correct that what happened is incorrect, it seems that it probably didn't start happening with Nero 11 and may happen in earlier, more perfect versions.

2. Just in case I never hear from tech support ...

Attached File(s)

 20120413_RecodeFLVStrangeness.zip ( 1.63MB ) Number of downloads: 81

 

[wither]

I didn't look at your files. However, there's nothing that says that the encoding is directly proportional to the precision settings (even though they're represented in even increments). It very well could be logrithmic. Let us know what tech support says.

[ArmandLiberi]

wither:

If you look at the files, you will see that the bit rate, whether linearly or logarithmicly, sometimes goes down while the precision goes up. That's the issue I'm addressing. Sorry for the confusion.

I did the same thing encoding 3GP output files and got similar results. I made a graph. Does it look correct?









This post has been edited by ArmandLiberi: Apr 14 2012, 05:34 AM

Attached thumbnail(s)

 

[Mister_M]

The resulting bit rate isn't really related to the encoder precision itself, but to its used encoding matrix. Depending on higher precision and therefore a possibly more efficient matrix, the picture quality can be as good as at lower precisions, but having also a lower bitrate at the same time.

[wither]

Well, that's clear to the uninitiated. Not! I'm wondering if the work "precision" is the correct one. Precision usually relates to reproducibility, not accuracy. It's not clear to me what the feature is actually doing. Maybe you can dummy your response down a little so that we can understand it. :D

[Mister_M]

Well, now it gets complicated... crusty in the Doom9's forum gave a good explanation for encoder matrices (also called quantizer or quantization matrices): http://forum.doom9.org/showthread.php?p=318375

Here you can see two examples, how a matrix can be read by an encoder.

But long story short: The picture quality depends not only on the bit rates but also on the efficiency of the used algorithm in the quantizer (encoding) matrix (but that's too much mathematics also for me!). A very well known algorithm is the Huffman algorithm.

This is the reason, why in higher precision modes lower bit rates with the same picture quality are possible.

As far as I remember you were able to import customized matrices (or edit the H.263 or MPEG matrix) in Nero Recode 2 (Nero 7).

[wither]

Now it's really perfectly clear-somewhat. :D As I understand it, values are assigned to some bits in the 8 x 8 matrix. Then the DCT calculates frequency values from those values and either replaces each bit in the matrix or some bits in the matrix. The matrix is then solved to get the value. The latter is where I got confused. It's been a long time since I got into that kind of stuff and I won't understand it by reading for 5 minutes.

What I understood the most was the basis for assigning bits for colors versus brilliance.

Thanks for helping on this.

[ArmandLiberi]

wither & Mister_M:

Thanks for the input. I guess I have some reading to do!

I did come across this lines in every Recode log file that I have where I didn't set the encoding precision:

EncodingPrecision: 400

(It appears to be a printing error!)


Do you have Recode log with a "EncodingPrecision: 400" line in it?


This post has been edited by ArmandLiberi: Apr 15 2012, 11:39 PM

[ArmandLiberi]

Not being a "true believer" as yet, I tried to verify the information I received when I posted this topic. To my satisfaction, I think I have done that. Here are the steps I took:

1. Convert the Butterfly_4_3.jpg sample to Butterfly_4_3.BMP using Microsoft Office Picture Manager.

2. Using one of the many schemes I found to do jpeg compression of varying quality, I made ten new .BMP files varying the "quality factor" from 10 to 100 in steps of 10. Since I did the DCT, quantification, and iDCT but not the actual compression, these files are all the same length but of varying quality.

3. I converted the files from .bmp to .jpg using Microsoft Office Picture Manager with a quality setting of 100. These files have lengths that increase as the quality of the input file increases. (As I initially thought they would/should).

4. I converted the files from .bmp to .jpg using Microsoft Office Picture Manager with a quality setting of 80. These files have lengths that do not necessarily increase as the quality of the input file increases. In other words, they exhibit the same "strangeness" I originally asked about.

Since I have no specifications for the MP4 or any other video format, that's about all I can do now (or care to do!). The one question that remains: How can a person verify that the "higher quality" output files actually have "higher quality"? After reading about all of the various color schemes, I don't even want to think about that.

The files are attached (temporarily). Thank you all for your input to this topic.


Note: Files can be downloaded from the public directory at

https://skydrive.live.com/?cid=4c028ef5ce7a...5CE7A1E35%21107#

There you will find PlayingWithBitRate.zip.

This post has been edited by ArmandLiberi: Jul 5 2012, 11:48 PM