GSoC 2013: Contour detection or feature detection ?

[Chintak Sheth]

Hi,

I am writing this mail since I am a little confused as to which project
would be the most advantageous to our community. After talking to Tony, I
have been able to short-list 2 projects, one regarding contour detection
using snakes or feature detection using SIFT. Kindly take a look at the
ideas described in detail below.

*About me:*
I am Chintak Sheth a 3rd year UG pursuing B.E. (Hons.) in Electrical and
Electronics engineering at BITS Pilani Goa campus. I have been working in
Image processing for about 8-9 months. I took it as an elective course just
out of curiosity, and am now really passionate to work in it, I have
recently got a paper published in IEEE Xplore DL. "Study of Dissociated
Dipoles for Content Based Image Retrieval Using Stamp Recognition
Systems<http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=6498274&contentType=Conference+Publications&searchField%3DSearch_All%26queryText%3Ddipole+histogram>"
At the time we had worked on MATLAB, though I regret not having got exposed
to skimage then. This is my first exposure with open source development.
But in the last couple of weeks, I have contributed to the community by way
of documentation for Morphological functions (with a lot of help from Tony:
https://github.com/scikit-image/scikit-image/pull/528) which was recently
merged and also by trying to implement a convex_hull_object function (
https://github.com/scikit-image/scikit-image/pull/522).

*Idea 1:*
I was fascinated about value-and-criterion class of nonlinear filters after
reading "A morphology-based filter structure for edge-enhancing
smoothing<http://ieeexplore.ieee.org/xpl/articleDetails.jsp?reload=true&arnumber=413627>".
It describes the implementation of Mean of Least Variance filters. What is
interesting is the edge preserving smoothing behaviour and also the ease of
extending it to 3D or n-D.

Stefan further suggested going through this paper (Edge and Line Oriented
Contour Detectors : State of the
Art<http://www.cs.rug.nl/~petkov/publications/2011ivc-contour_survey.pdf>)
which deals with a higher domain of contour detection and also presents on
the various stages involved in this. As it turns out value-and-criterion
filters fall into the preprocessing stage for contour detection and are
mainly used for removing noise and textures while at the same time
preserving edges. For the detection part, I think Active contours (or
snakes) offers some pretty serious advantages as opposed to local
detectors. Specifically, Snakes: Active contour
models<http://morse.cs.byu.edu/750/papers/kass-witkin-terzopoulos-snakes.pdf>
describes
the basic algorithm and has been cited about 14660 times.

I think snakes in combination with a preprocessing filter would be a pretty
good addition to skimage.

*Idea 2:*
A feature detector like SIFT. A lot of literature is available online and
also should be pretty good to have in skimage.

I'd love to know what are your thoughts on what could be more advantageous
to our community.

Thanks,
Chintak
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-image/attachments/20130427/c4717582/attachment.html>