[Mp4-tech] MPEG4 Simple Profile question. Is DC/AC saturation in the range [-2048, 2047] redundant provided that bits_per_pixel is 8?

[Gary Sullivan]

Shevach,
That note may impose a limit on something called dct_dc_size_luminance, but but I am fairly sure that it does not impose a limit on the value obtained by DC prediction.
Let me pose a hypothetical case for you.
Let's say that an encoder encodes a picture such that the value of the encoded residual for every DC coefficient in the entire picture is 128.  If the width of the picture is Y in units of DCT blocks, then in the absence of clipping I believe the value of the DC coefficient of the right-most DCT block in the top row will be approximately 128*Y.  So, for example, if the picture has CIF resolution, the DC coefficient of the right-most block will become approximately 5632, which is bigger than 2047.
And now think about the value of the DC coefficient of the right-most DCT block in the bottom row of the picture.
Best Regards,
Gary Sullivan
________________________________
From: Shevach Riabtsev [mailto:sriabtsev broadcom.com]
Sent: Thursday, April 17, 2024 9:28 AM
To: Gary Sullivan; mp4-tech lists.mpegif.org
Subject: RE: [Mp4-tech] MPEG4 Simple Profile question. Is DC/AC saturation in the range [-2048, 2047] redundant provided that bits_per_pixel is 8?
Gary
Thanks for your reply. I completely agree with you but one subtle point.
In Simple Profile DC residuals are limited to the range [-511,511] according to the Note in B.1.4 and Table B-15:
The variable length code for dct_dc_size_luminance of 10, 11 and 12 are not valid for any object types where the pixel depth is 8 bits. They shall not be present in a bitstream conforming to these object types.
Therefore DC values obtained from DC prediction will not exceed [-1024..1024] range. So, I see that the DC clipping is redundant in a Simple Profile decoder.
Regards
________________________________
From: Gary Sullivan [mailto:garysull windows.microsoft.com]
Sent: Thursday, April 17, 2024 2:47 AM
To: Shevach Riabtsev; mp4-tech lists.mpegif.org
Subject: RE: [Mp4-tech] MPEG4 Simple Profile question. Is DC/AC saturation in the range [-2048, 2047] redundant provided that bits_per_pixel is 8?
Shevach Riabtsev et al,
I think the clipping is necessary.  The first thing to remember is that the standard specifies the decoding process and the constraints on bitstream conformance.  It does not specify how encoding is done.  So if you are trying to base your rationale on what you assume that an encoder must be doing, you need to give up and stop looking at things from that perspective.
Section A.1 was replaced a year or two ago - part of the purpose of that replacement was to clarify what the standard was trying to say.  It would be best to make sure that you're looking at the latest version of that section if you want to refer to something in it.  And make sure that you're aware of the difference between normative and non-normative aspects of the standard.  My guess is that you're trying to refer to the forward DCT operation, which is not normative, since that is something that happens (if it happens) only in an encoder - and encoding methods are not normatively specified by the standard.  Encoders are not actually required to use a forward DCT.
I think you are correct that dc_scaler is always at least 8 for Simple profile.
There is an equation at the beginning of 7.4.3.2 that specifies how QFx[0][0] is computed using PQFx[0][0], dc_scaler, and Fa[0][0] or Fc[0][0].  So to determine the range of values, prior to the clipping specified at the beginning of 7.4.3.4, you should examine the ranges of values that those four variables can have, and then assume an "evil" worst case combination of them for the equation in 7.4.3.2.  Again, don't assume that the encoder is doing anything friendly.  Assume that the encoder has been designed to try to force a violation of the clipping range without regard to whether you would consider what it is doing to be sensible.
Best Regards,
Gary Sullivan
________________________________
From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Shevach Riabtsev
Sent: Wednesday, April 16, 2024 6:55 AM
To: mp4-tech lists.mpegif.org
Subject: [Mp4-tech] MPEG4 Simple Profile question. Is DC/AC saturation in the range [-2048, 2047] redundant provided that bits_per_pixel is 8?
Dear experts
I have a question on MPEG4 Simple Profile related to DC/AC quantized coefficient saturation.
According to 7.4.3.4
The quantized coefficients resulting from the DC and AC Prediction are saturated to lie in the range [*2048, 2047].
In MPEG4 Simple Profile bits_per_pixel value is 8, I don't see a case when a quantized DC or AC coefficient can exceed [-2048,2047] range.
For example,  DC obtains the maximal value equal 8*255 = 2040 provided that all pixels in a 8x8 block are 255, see A.1. The minimal DC quantization scaler is 8 (see Table 7-1). Thus the maximal quantized DC magnitude is 2040/8=255. Hence DC residual lies in the interval [-255..255].  The same reasoning can be applied to AC case.
Thus I conjecture that AC/DC coefficients satuaration is redundant in Simple Profile. Am I right?
Regards, Shevach
Broadcom
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20080417/91834e0a/attachment-0001.html