Posted by
mahogny on
URL: http://confocal-microscopy-list.275.s1.nabble.com/Re-Ana-alarming-amount-of-image-manipulation-tp592857p1313586.html
Martin Wessendorf wrote:
> Dear Johan--
>
> Johan Henriksson wrote:
>
>> compression does NOT imply reduced image quality
> >
>> but it does for sure increase the speed of disk transfer, and if you
>> know what you are doing, some but not all compression algorithms can
>> *optionally* trade high frequency information (noise) for disk space.
>> avoid using the terminology that "compression" destroys images because
>> it confuses non-experts into thinking they should be storing the images
>> uncompressed. there's a factor 2-10 to gain in disk space/speed for
>> normal images.
>
> Could you please elaborate about storing images compressed? Are you
> talking about lossless compression or (as it sounds like from the
> context) lossy compression?
I can sum up the current situation
======
there are 3 major open formats in use now:
TIFF. uncompressed or lossy JPEG. I think there is a rather bad lossless
format included that essentially no one supports
JPEG. lossy
PNG. lossless
TIFF supports 16bit but is rather awful overall. I recommend against it
whenever possible, because it is loosely specified. the result is that
some programs write images other programs cannot read. the problem with
JPEG and PNG is that they only support 8-bit images as far as I know,
and there are no real lossless formats competing with PNG. I would be
happy to see that someone extended PNG, the current algorithm with just
a larger word size would do the trick.
my benchmarks show that fluorescent images compress extremely well with
the PNG lossless algorithm so the only valid excuse not to use it at the
moment is if you have 12 or 16 bits. about a factor 10 in compression.
PNG does not work well on DIC images, about a factor 2.
JPEG, or any lossy compression, always wins over lossless since it in
practice applies a lowpass filter before storing down the data. JPEG
gives a factor 10 or more on DIC images, depending on your tradeoff.
======
in our lab, we use lossy compression (JPEG) on DIC images and lossless
(PNG) on fluorescent. motivation:
* DIC images give a qualitative impression, and as long as a human can
see what is there, we're fine.
* fluorescent images are used in quantification, and you have to be
careful with removing information.
* we cannot store all images with just lossless compression. we have
over 100 000 000 image planes, closing in on 2TB data.
under normal conditions we add 6GB data per day with the above scheme
(time lapse), as opposed to 50GB before I improved the setup. the
alternative to save space is to not capture an image at all, but I
consider this to be by far worse for research than storing images
without the noise.
======
the future: I would like to see some applied computer scientists work on
standardizing new formats. there is for example JPEG2000 which is by far
better than JPEG (better base functions), but cannot really be used
since no one implements it. likewise, it would be trivial to extend PNG
to higher pixel depths. but given that we now work with 3D datasets, or
even 5D, lossless compression rates can be improved even further, so
this should really be investigated.
there are more issues to be resolved. any format that stores all images
to one single file has all sorts of issues. some formats even store
images as XML, I won't even get started about these. then there is the
whole metadata issue. we started our own format standardization
initiative (for OST) since no format delivered what our lab required.
/Johan
--
--
------------------------------------------------
Johan Henriksson
MSc Engineering
PhD student, Karolinska Institutet
http://mahogny.areta.org http://www.endrov.net