A new asymmetrical corner detector(ACD) for a semi-automatic image co-registration scheme

摘要

Co-registration of multi-sensor and multi-temporal images is essential for remoteudsensing applications. In the image co-registration process, automatic Ground ControludPoints (GCPs) selection is a key technical issue and the accuracy of GCPs localizationudlargely accounts for the final image co-registration accuracy. In this thesis, a noveludAsymmetrical Corner Detector (ACD) algorithm based on auto-correlation isudpresented and a semi-automatic image co-registration scheme is proposed.udThe ACD is designed with the consideration of the fact that asymmetrical cornerudpoints are the most common reality in remotely sensed imagery data. The ACD selectsudpoints more favourable to asymmetrical points rather than symmetrical points to avoidudincorrect selection of flat points which are often highly symmetrical. The experimentaludresults using images taken by different sensors indicate that the ACD has obtainedudexcellent performance in terms of point localization and computation efficiency. It isudmore capable of selecting high quality GCPs than some well established corneruddetectors favourable to symmetrical corner points such as the Harris Corner Detectorud(Harris and Stephens, 1988).udA semi-automatic image co-registration scheme is then proposed, which employs theudACD algorithm to extract evenly distributed GCPs across the overlapped area in theudreference image. The scheme uses three manually selected pairs of GCPs to determineudthe initial transformation model and the overlapped area. Grid-control and nonmaximumudsuppression methods are used to secure the high quality and spreaduddistribution of GCPs selected. It also involves the FNCC (fast normalised crosscorrelation)udalgorithm (Lewis, 1995) to refine the corresponding point locations in theudinput image and thus the GCPs are semi-automatically selected to proceed to theudpolynomial fitting image rectification. The performance of the proposed coregistrationudscheme has been demonstrated by registering multi-temporal, multi-sensorudand multi-resolution images taken by Landsat TM, ETM+ and SPOT sensors.udExperimental results show that consistent high registration accuracy of less than 0.7udpixels RMSE has been achieved.udKeywords: Asymmetrical corner points, image co-registration, ACD