[Mp4-tech] [video] - transform & quantization

Sat Aug 14 21:30:17 EDT 2004

really good reply
veni
enJOY life
-----Original Message-----
From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Gary Sullivan
Sent: Friday, August 13, 2023 11:41 PM
To: Tommaso fambau.it; mp4-tech lists.mpegif.org
Cc: wien ient.rwth-aachen.de
Subject: RE: [Mp4-tech] [video] - transform & quantization
Tommaso,
Response in-line below: 
+> -----Original Message-----
+> From: mp4-tech-bounces lists.mpegif.org
+> [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of 
+> Tommaso fambau.it
+> Sent: Friday, August 13, 2023 7:32 AM
+> To: mp4-tech lists.mpegif.org
+> Subject: [Mp4-tech] [video] - transform & quantization
+> 
+> 
+> Hi,
+> i'm a student of university of Padua.
I hope you are enjoying your student life, and I am happy to see that you are digging deeply into your studies.
+> I've just started a study on the trasform and quantization designs in 
+> H.264/AVC and there are some questions i'm not been able to answer 
+> also searching on the net. I hope you would help me.
+> 
+> I start saying that i've based my studies on the IEEE
+> TRANSACTIONS (july
+> 2003) and in particular mode on the papers  "low-complexity 
+> transform and
+> quantization in h.264/avc"  and  "variable block size transform for
+> h.264/avc".
There is also some tutorial information on this subject that is available on the web from Iain Richardson.  Google or MSN Search can help you find it (a little digging with a good search engine can find all sorts of stuff).  He also wrote a book, with similar content.
You should make sure to also get a copy of the standard.  In particular, since you are interested in the variable block-size capability, you should also get a copy of the draft FRExt amendment.  When you get into the details, you should study the standard more closely than you study the papers about the standard.
Especially keep in mind the distinction between what is standardized (the decoding method including the inverse transform and its scaling) and what is not (the encoding method including the forward transform and quantization).
If you want to become a real video coding stud, figure out every detail of exactly how SP and SI pictures work.
+> The questions are the next (i write a list to make it easier):
+> 
+> 1- what's DYNAMIC RANGE of a block?
That is the range of values that can occur during processing of the input data in an encoder or decoder. For example, if the decoded video has 8 bits per sample, the dynamic range of a decoded sample is 8 bits, or 0 to 255. What is coded is the difference between the actual value of a sample (which ranges from 0 to 255) and the prediction of the value of the sample (which also ranges from 0 to 255), so the input to the forward transform has a dynamic range from -255 to +255.  Then when you consider the possible combinations of values in a 4x4 block or a larger block, the dynamic range of a transform coefficient has another dynamic range, etc.
+> 2- Are the TRANSFORM DC COEFFICIENTS simply the coefficients
+> in the first
+> plece at the top-left of the block (so they are the continuous
+> coefficients)?
The "DC" coefficient is the coefficient with coordinates (0,0) in the transform domain.  It represents the average value of the block of samples in the spatial domain.  In the forward transform, it is just the scaled sum of all input sample values.  It ended up getting called "DC" because it represents the part of the signal that is not varying.
+> 3- The standard describe the 4x4 and 8x8 transforms. How 
+> could the 2x2
+> blocks of DC coefficients of chrominance be trasformed 
+> (pag.601 if IEEE
+> TRANSACTIONS)??
The 2x2 blocks of DC coefficients for chroma are transformed by a simple "Hadamard" transform.  In other words, a simple sum and difference are created both horizontally and vertically.
+> 4- What are the QUANTIZATION TABLES? Could i find them some where?
They are tables of frequency-dependent scaling multipliers.  They can have fixed values, as in the original version of H.264/AVC, or they can be encoder-specified in some fashion, as in FRExt.
+> 5- On the paper "low-complexity..." pag.602, is written: "The basis
+> function of the new transform do not have equal norm, which 
+> leads to an
+> increade in the size of quantization tables". why this 
+> increase?? i can't
+> understand!
The scaling multipliers become frequency-dependent when the basis functions of the transform do not have equal norms.  So instead of needing one table that applies to all coefficients, you need several different tables, where the table that you use depends on the norm of the associated basis function.
+> 6- On the paper "variable..." pag.605, the 8x8 transform 
+> matrix is shown,
+> and is written that is a SINGLE NORM matrix. From my 
+> calculations on the
+> rows i found it isn't so. Where is my mistake??
I haven't checked it.  Perhaps you should contact the author (who I have copied on this message).  In any case, a different 8x8 transform was eventually used in FRExt than what was described in that paper.  The one in FRExt is not a single-norm matrix.
+> I hope i've been clear and i excuse me for so much questions.
If each question leads to an equal increment of understanding, more questions are better than fewer questions.
Best Regards,
Gary Sullivan
+> 
+> thanks.
+> 
+> Tommaso Baù
+> 
+> 
+> _______________________________________________
+> NOTE: Please use clear subject lines for your posts. Include 
+> [audio, [video], [systems], [general] or another 
+> apppropriate identifier to indicate the type of question you have.
+> 
+> Note: Conduct on the mailing list is subject to the 
+> Antitrust guidelines found at 
+> http://www.mpegif.org/public/documents/vault/mp-out-30042-Ant
+> itrust.php
+> 
_______________________________________________
NOTE: Please use clear subject lines for your posts. Include [audio, [video], [systems], [general] or another apppropriate identifier to indicate the type of question you have.
Note: Conduct on the mailing list is subject to the Antitrust guidelines found at http://www.mpegif.org/public/documents/vault/mp-out-30042-Antitrust.php