W3C home > Mailing lists > Public > public-respimg@w3.org > September 2013

Re: Progressive JPEG (was: Compressive images test)

From: Frédéric Kayser <f.kayser@free.fr>
Date: Wed, 18 Sep 2013 03:49:01 +0200
To: public-respimg@w3.org
Message-Id: <D7501835-68FF-4C83-919D-57FF070A91A9@free.fr>
I've turned the unicorn into a 7 scans progressive JPEG (spectral selection only) and made an animated GIF (and APNG) out of it:
http://frdx.free.fr/unicorn/
(the alt version is a bit different)

The top data units hold what the decoder has at hand to reconstruct the image at the end of the current scan, the bottom ones are the encoder point of view (a red triangle shows a coefficient that is currently written).

In the first scan only the DC coefficient (upper left, the most important) of each data unit is written, all the scans are interleaved since the 3 components are kept together (the alt version differs on this point).
In the second scan the 2 first AC coefficients (following the zigzag path) of each data unit are written.
in the third, fourth and fifth scan respectively 3, 4 and 5 AC coefficients of each data unit are written.
in the sixth scan 13 AC coefficients of each data unit are written.
in the last scan all the remaining 36 AC coefficients of each data unit are written.

In a sequential JPEG entire data units would have been written to the file one of each component (no chroma subsampling here) at a time, it's easier to decode and there's no need to keep the partially filled data units in memory.


Regards
-- 
Frédéric Kayser

Frédéric Kayser wrote :

> I'll try to turn the Unicorn in real illustration of this process.



unicorn-progressive.png
(image/png attachment: unicorn-progressive.png)

Received on Wednesday, 18 September 2013 01:49:52 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 20:12:40 UTC