Conclusions
In this paper, an accurate affine invariant image matching ap-
proach is introduced for high accurate feature correspondence
in image pairs with significant geometric differences. The
proposed algorithm starts by detecting local affine invariant
features in an image pair using an integrated uniform robust
approach based on
MSER
and Harris-Affine algorithms. Then,
initial feature correspondence and mismatch elimination
process are performed using
SIFT
feature descriptor compari-
son and a consistency check process. Finally, an advanced
oriented least square matching (
OLSM
) method is developed
to obtain a set of high accurate corresponding features. The
proposed
OLSM
method is a novel extension of the standard
LSM
that is provided to perform an accurate affine invariant
image matching with significant local distortion. Quantitative
evaluations show that the proposed approach outperforms
standard
LSM
technique both in terms of accuracy and success
rate for both synthetic and real inter-band images.
The proposed
OLSM
matching approach can be applied to
any similar affine invariant feature extraction algorithms such
Figure 8. Positional accuracy (
rmse
) results on the inter-band simulated image pairs for scale (left column), rotation (middle column), and
viewpoint (right column) geometric distortion: (a) first image:
ubc
, (b) second image: Graffiti, (c) third image:
spot5
, and (d) forth image:
Worldview
740
September 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
