[scikit-image] Nonrigid image registration in Python

[Jérôme Kieffer]

On Fri, 07 Oct 2016 14:15:21 -0700
Stefan van der Walt <stefanv at berkeley.edu> wrote:

> On Thu, Oct 6, 2016, at 12:29, Jérôme Kieffer wrote:
> > We have implemented the SIFT algorithm on GPU a few years ago.
> > It is now part of the silx toolkit
> > https://github.com/silx-kit/silx/blob/master/doc/source/Tutorials/Sift/sift.ipynb
> 
> Note that SIFT is patent encumbered, that is why we have CENSURE
> features in skimage.

Hi Stephan,

I know about the patent, but:
1. I live in Europe, where it is not applicable
2. I have neither a passport nor a visa to the USA (where it is applicable)
3. The patent is quite old and should expire by a couple of years from now.

When we did the study (i.e. 2011-2012), SURF was claimed "patent-free"
and this no more the case now :( I wish you the same story does not
apply to CENSURE.

The difference of boxes is similar to SURF, CENSURE is just using octogons,
another trick to avoid the patent on gaussian blurring. By the way the
difference of boxes has been patented with SURF if I remember well.

My question would rather be about the quality of the descriptors ... 
in skimage you provide BRIEF, does this mean CENSURE has no specific
keypoint descriptor ?

Cheers,

Jérôme