[IPOL discuss] 16 bits images

Nicolas Limare nicolas.limare at cmla.ens-cachan.fr
Mon Mar 5 10:49:46 CET 2012

Hi Miguel and everyone,

> currently the IPOL conversion mechanism is converting the input images to
> 8 bits before processing them with the algorithms.

Yes, when someone uploads an image file, it is converted to the file
format required for the algo program. For example, when the program
needs a PNG grayscale input, and someone uploads a JPEG, TIFF or GIF,
this uploaded image is converted to PNG grayscale. This conversion is
only performed as a pre-processing, before the input file is made
available for the program.

> Unfortunately, some applications (like noise estimation or denoising,
> among others) need almost noise-free images as their input.
> When encoded with the PNG format, a quantification error is added, and it
> interferes with the noise in the image.

Which input format would not be quantized? Are you refering to
floating-point TIFF files? Does anyone have floating-point TIFF
noise-free image files to submit to an IPOL demo? And if the
quantization is an issue, then the gamma encoding is another one, and
the input preprocessing should also take care of converting any input
image to linear encoding.

This makes me think that, once again, we may be asking too much to
the demo system. It is only a demo, not a networked interactive
computing environment. It is inherently limited by the web
technologies and the amount of fine tuning we are able to invest in
its development.

Creating a complete full-featured web-based experimental environment
intended to replace what we currently do on our workstations would
require a much larger development effort and workforce, and is by
itself a major research project, independent of the IPOL "reproducible
computational science publishing with demo" initiative.

> This quantification error can't be avoided, but it can be minimized is the
> images are encoded with 16 bits instead of 8 bits.
> My opinion is that IPOL should implement the 16 bits conversion as soon as
> possible, because the quantification error in the input images is too high
> for many applications.

I agree, it would be good to be able to specify the demos with 16bit
input image files. And I have the same opinion about your previous
(unaswered) suggestion to have different processing methods for color and
grayscale images. I do not have much time now to undertake this task,
and not much motivation for web app development either. But any patch
to implement these enhancements to the demo system (without breaking
existing demos) is welcome.


Nicolas LIMARE - CMLA - ENS Cachan    http://www.cmla.ens-cachan.fr/~limare/
IPOL - image processing on line                          http://www.ipol.im/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://tools.ipol.im/mailman/archive/discuss/attachments/20120305/e5292f26/attachment.pgp>

More information about the discuss mailing list