Hi,
To achieve best results with compression it is important to choose the
best input to compress. That is, given a source image, for example
1920x1080 4:2:2, you can compress it directly, or you could convert it
to a lower resolution such as 1280x720 4:2:0 and then compress the lower
resolution. When you reduce the resolution of the imput picture before
compression you will reduce its quality becuase it will be slightly more
blurred than the original. However you make it easier for the codec to
compress it because there is less data to compress. This results in
fewer compression artefacts. An ideal codec would produce the same
output quality whatever the input resolution. However today's codecs are
not ideal. So sometimes it is better to reduce the resolution before
coding becuase the reduction in quality due to reduced resoution is less
that the impairment due to compression artefacts from trying to compress
the original resolution. For example for the BBC HD transmissions in the
UK we have broadcast 1440x1080 images rather than the full 1920x1080
resolution, because the final quality was better. Similar experiments
for internet streaming indicate that streaming 1280x720, at the bit
rates available, yeilds better quality than trying to compress the full
1920x1080. For VC-2 we also compressed 1440x1080 images for compression
of HD or SD SDI links (SMPTE RP 2047-3:2011).
The need to select the correct image resolution prior to compression is
particularly important when the bit rate is highly constrained, that is
when you are trying to achieve the highest compression ratio. In this
circumstance the quality of the coded image may reduce significantly for
even a small increase in data rate (the differential of quality versus
bit rate is high). So by reducing the bit rate by a fairly small amount
(e.g. by coding 4:2:0 rather than 4:2:2) you may a significant decrease
in encoding artefacts for only a small redcution in resolution. I
suspect this is what you are finding when you report that the quality
from coding 1280x720 8 bit 420 is better than when coding 1280x720 8 bit
422. The relative quality from coding these two formats will vary
depending on the bit rate. At higher bit rates 4:2:2 may look better
whereas at lower bit rates 4:2:0 may be better.
The trade off between resolution and coding artefacts is important.
However it is difficult to analyse using a PSNR metric because the
subjective effects of low of resolution (i.e. blurring) is different
from the subjective effect of compression artefacts. PSNR does not take
this into account, which makes it poor metric for such an analysis. For
this reason, or perhaps because this issue is not well understood, there
has not been much research into this. I don't know of any academic
papers that really address this issue.
Tim Borer
On 19/03/2014 00:41, yumeiwang_2000 wrote:
Hi,
The last mail can't be sent because of the size of attachment too big.
i delete one attachment and send again.
I have another question, i tried 1280x720 8 bit 422 and 420, with same
compress ratio, the 420 quality is much better than 422. Is there
something wrong?
------------------------------------------------------------------
发件人：yumeiwang_2000 <yumeiwang_2...@aliyun.com>
发送时间：2014年3月18日(星期二) 13:19
收件人：schrodinger-devel <schrodinger-devel@lists.sourceforge.net>
主 题：schrodinger 1.0.11 10bit
Hi,
I try to use schrodinger 1.0.11 to compress picture, i test 8bit mode,
it works well. However, i can't get correct picture with 10 bit mode.
The attachment is the original picture(1280x720 10bit 4:2:2 mode), but
after encode and decode, the size is 1843200 which should be 8 bit size.
i changed the encode.c for the test, could you please help me check
how i set the parameters?
Best Regards,
Anna
------------------------------------------------------------------------------
Learn Graph Databases - Download FREE O'Reilly Book
"Graph Databases" is the definitive new guide to graph databases and their
applications. Written by three acclaimed leaders in the field,
this first edition is now available. Download your free book today!
http://p.sf.net/sfu/13534_NeoTech
_______________________________________________
Schrodinger-devel mailing list
Schrodinger-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/schrodinger-devel
------------------------------------------------------------------------------
Learn Graph Databases - Download FREE O'Reilly Book
"Graph Databases" is the definitive new guide to graph databases and their
applications. Written by three acclaimed leaders in the field,
this first edition is now available. Download your free book today!
http://p.sf.net/sfu/13534_NeoTech
_______________________________________________
Schrodinger-devel mailing list
Schrodinger-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/schrodinger-devel
