Co-registration of multi-sensor and multi-temporal images is essential for remote
sensing applications. In the image co-registration process, automatic Ground Control
Points (GCPs) selection is a key technical issue and the accuracy of GCPs localization
largely accounts for the final image co-registration accuracy. In this thesis, a novel
Asymmetrical Corner Detector (ACD) algorithm based on auto-correlation is
presented and a semi-automatic image co-registration scheme is proposed.
The ACD is designed with the consideration of the fact that asymmetrical corner
points are the most common reality in remotely sensed imagery data. The ACD selects
points more favourable to asymmetrical points rather than symmetrical points to avoid
incorrect selection of flat points which are often highly symmetrical. The experimental
results using images taken by different sensors indicate that the ACD has obtained
excellent performance in terms of point localization and computation efficiency. It is
more capable of selecting high quality GCPs than some well established corner
detectors favourable to symmetrical corner points such as the Harris Corner Detector
(Harris and Stephens, 1988).
A semi-automatic image co-registration scheme is then proposed, which employs the
ACD algorithm to extract evenly distributed GCPs across the overlapped area in the
reference image. The scheme uses three manually selected pairs of GCPs to determine
the initial transformation model and the overlapped area. Grid-control and nonmaximum
suppression methods are used to secure the high quality and spread
distribution of GCPs selected. It also involves the FNCC (fast normalised crosscorrelation)
algorithm (Lewis, 1995) to refine the corresponding point locations in the
input image and thus the GCPs are semi-automatically selected to proceed to the
polynomial fitting image rectification. The performance of the proposed coregistration
scheme has been demonstrated by registering multi-temporal, multi-sensor
and multi-resolution images taken by Landsat TM, ETM+ and SPOT sensors.
Experimental results show that consistent high registration accuracy of less than 0.7
pixels RMSE has been achieved.
Keywords: Asymmetrical corner points, image co-registration, ACD