On Sun, Apr 28, 2013 at 2:43 AM, Kyle Banerjee <[log in to unmask]>wrote:
Every choice people make is about loss. Equipment, optics, lighting,
> you name it. But for some reason, the instant we're talking about bits of
> on a disk, people plan as though capacity were unlimited when most
> archives are severely underresourced.
Strictly speaking, it is not correct to say that every choice is about
loss (or cost), and for once I'm saying this in a case where the difference
is actually significant. [Someone help edsu to the fainting couch.]
If a particular set of choices are below the Production Possibility
then those choices are strictly inferior to those that are on the Frontier.
Why is this relevant? Because, for the situation where lossless image
storage has a very value, TIFF is not the most space efficient way of
storing the data.
A month or so ago I did a few measurements, using a (not necessarily
representative) color photograph (TIFF extracted from a Canon EOS-10D raw).
For lossless conversion, I used uncompressed TIFF, compressed TIFF, PNG,
and JP2 (100% quality). Measurements using the ImageMagick "compare"
utility confirmed zero signal loss:
-rw-r--r--@ 1 ses staff 18M Mar 19 14:52
-rw-r--r--@ 1 ses staff 9.4M Mar 19 14:53 CRW_4237_tiff_8_compressed.tif
-rw-r--r-- 1 ses staff 8.2M Mar 19 14:29 CRW_4237-0.png
-rw-r--r--@ 1 ses staff 6.1M Mar 19 14:03 CRW_4237_quality_100-0.jp2
For lossy compression, using RMSE as the metric, we can see that JPEG at
90% quality is showing measurable signal degradation, with a compression
ratio of 4.7:1 relative to the JP2 file (vs. 14:1 relative to uncompressed
tiff, and 7.2:1 for compressed).
$compare ... CRW_4237_jpg_90.jpg = 459.806 (0.00701619) [1.3M]
JP2 at quality 75 showed slightly less signal loss by RMSE, with a
compression ratio of 5.5 : 1
$compare ... CRW_4237_quality_75.jp2 = 457.959 (0.006988) [1.1M]
Note that the image type was a color photograph; other image types may get
better lossless compression using PNG or TIFF. Also, some people have
expressed concern over the use of JP2 for archival purposes due to a
relatively small number of open-source libraries. On the other hand, JP2
has some potentially useful properties for distributed replicated
preservation (layers with fine levels of detail could be split off and
stored on fewer replicas).