Invariant Features from
Interest Point Groups
Matthew Brown and David Lowe
{mbrown|lowe}@cs.ubc.ca
Department of Computer Science,
University of British Columbia,
Vancouver, Canada.
Abstract
This paper approaches the problem of finding correspondences b etween
images in which there are large changes in viewpoint, scale and illumi-
nation. Recent work has shown that scale-space ‘interest points’ may
be found with good repeatability in spite of such changes. Further-
more, the high entropy of the surrounding image regions means that
local descriptors are highly discriminative for matching. For descrip-
tors at interest points to be robustly matched between images, they
must be as far as possible invariant to the imaging process.
In this work we introduce a family of features which use groups
of interest points to form geometrically invariant descriptors of image
regions. Feature descriptors are formed by resampling the image rel-
ative to canonical frames defined by the points. In addition to robust
matching, a key advantage of this approach is that each match implies
a hypothesis of the local 2D (projective) transformation. This allows
us to immediately reject most of the false matches using a Hough trans-
form. We reject remaining outliers using RANSAC and the epipolar
constraint. Results show that dense feature matching can be achieved
in a few seconds of computation on 1GHz Pentium III machines.
1 Introduction
A widely-used approach for finding corresp onding points between images is to de-
tect corners and match them using correlation, using the epipolar geometry as a
consistency constraint [3, 13]. This sort of scheme works well for small motion,
but will fail if there are large scale or viewpoint changes between the images. This
is because the corner detectors used are not scale-invariant, and the correlation
measures are not invariant to viewpoint, scale and illumination change. The first
problem is addressed by scale-space theory, which has proposed feature detec-
tors with automatic scale selection [6]. In particular, scale-space interest point
detectors have been shown to have much greater repeatability than their fixed
scale equivalents [7, 8]. The second problem (inadequacy of correlation) suggests
the need for local descriptors of image regions that are invariant to the imaging
process.
评论0