[Image-SIG] feature detection/location in images

Chris Ps lubensch.proletariat.inc at gmail.com
Tue Jul 14 19:05:54 CEST 2009


> ---------- Forwarded message ----------
> From: Kevin Cazabon <kevin at cazabon.com>
> To: image-sig at python.org
> Date: Mon, 13 Jul 2009 23:31:14 -0400
> Subject: [Image-SIG] feature detection/location in images
> Hey everyone - does anyone have good experience in locating features in an
> image?  PilPlus used to have a "ImageCrackCode" module, but I've never used
> it or even seen it (it says no longer available).
>
> I don't need to find _specific_ features, but need to find a small set of
> distinct features consistently in similar images (for aligning them
> together).  I can brute-force this with min/max values/etc, but was hoping
> for some help on something more elegant and efficient.  Any pointers would
> be appreciated, thanks!
>
> Kevin.
>
>
>
> ---------- Forwarded message ----------
> From: "Guy K. Kloss" <g.kloss at massey.ac.nz>
> To: image-sig at python.org
> Date: Tue, 14 Jul 2009 16:37:21 +1200
> Subject: Re: [Image-SIG] feature detection/location in images
> On Tue, 14 Jul 2009 15:31:14 Kevin Cazabon wrote:
> > I don't need to find specific features, but need to find a small set
> > of distinct features consistently in similar images (for aligning them
> > together).  I can brute-force this with min/max values/etc, but was
> > hoping for some help on something more elegant and efficient.  Any
> > pointers would be appreciated, thanks!
>
> I've accomplished something similar using OpenCV functions that correlate a
> template to an area and return a matrix of correlation ratings. Then
> finding
> in the matrix the extreme value (max or min depending on the function) you
> can
> find your "best match".
>
> Currently I'd advise the ctypes-opencv bindings by Minh-Tri Pham to use the
> code from Python. It's quite speedy and works well, just need to play a bit
> with the functions to find out which works best for your case.
>
> HTH,
>
> Guy
>
> --
> Guy K. Kloss
> Institute of Information and Mathematical Sciences
> Te Kura Pūtaiao o Mōhiohio me Pāngarau
> Massey University, Albany (North Shore City, Auckland)
> 473 State Highway 17, Gate 1, Mailroom, Quad B Building
> voice: +64 9 414-0800 ext. 9585   fax: +64 9 441-8181
> G.Kloss at massey.ac.nz http://www.massey.ac.nz/~gkloss<http://www.massey.ac.nz/%7Egkloss>
>

I would also recommend OpenCV. I'm not familiar with the bindings by
Minh-Tri Pham, I'm using the Python(x,y) distribution which includes python
bindings from the original c library. Some of the functions are buggy and
the API is rather clumsy (very c like), but I don't think you will have a
problem. In case the functions you are looking for don't exist in OpenCV a
solution would be to export your PIL image as a nd array to numpy and work
with that. Dealing with numpy makes repetitive window based operations
faster than PIL.

Christos
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/image-sig/attachments/20090714/473f6383/attachment-0001.htm>


More information about the Image-SIG mailing list