[OctDev] SIFT image descriptor patch

[Etienne G. <et...@cs...>]

  Hi All,

this patch could be of interest to octave users who do image
processing.  You can apply it to the matlab code in siftDemoV4 [1] to
allow octave to extract SIFT image features [2] with David Lowe's
package.

  The SIFT image detector identifies interest points, somewhat like a
corner detector does, and gives them a local descriptor. The
descriptor can then be used for matching purposes (another way to
perform matching could be correlation of a small window around the
feature). Wrt to corners + local window correlation, SIFT features are
more robust to changes in illumination and viewpoint [3] and they are
increasingly used in computer vision.

  The siftDemo has functions to find sift features and to perform
matching. I've tested the patch under octave+linux, matlab+linux (it
works) and octave+win (it fails, probs w/ imagemagick + double).

  Hth,

  Etienne



[1] http://www.cs.ubc.ca/spider/lowe/keypoints/siftDemoV4.zip

  Synopsis:=20

    unzip siftDemoV4.zip;
    cd siftDemoV4;=20
    patch -p1 < mySift.patch
 =20
    In octave, make sure sift.m is in your path, then

    >> [image, descriptors, locs] =3D sift("/home/etienne/prog/various/si=
ftDemoV4/book.pgm");
   >> showkeys(image, locs)

   See also 'help match'

[2] http://www.cs.ubc.ca/~lowe/keypoints/

  David G. Lowe, =D6bject recognition from local scale-invariant
  features," International Conference on Computer Vision, Corfu,
  Greece (September 1999), pp. 1150-1157.

[3] A performance evaluation of local descriptors Krystian
    Mikolajczyk, Cordelia Schmid International Conference on Computer
    Vision & Pattern Recognition - June 2003

    http://lear.inrialpes.fr/pubs/2003/MS03/mikolajczyk_cvpr2003.pdf


--=20
Etienne Grossmann ------ http://www.cs.uky.edu/~etienne