 |
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
|||||||||||
 |
 |
 |
|
|||||||||
 |
|
|||||||||||
 |
 |
 |
|
|||||||||
 |
|
|||||||||||
 |
 |
 |
|
|||||||||
 |
|
|||||||||||
 |
 |
|
||||||||||
 |
|
|||||||||||
 |
 |
 |
|
|||||||||
 |
|
|||||||||||
 |
 |
 |
|
|||||||||
 |
|
|||||||||||
 |
 |
 |
|
|||||||||
 |
|
|||||||||||
 |
|
|||||||||||
 |
|
|||||||||||
Automated Multisensor and Multitemporal Image Registration
Funded by the NOAA Coastal Change Analysis Program (C-CAP)
X. Long Dai and Siamak Khorram
OBJECTIVE
This research is to explore, develop, and implement automated algorithms for multisource remotly sensed data registration.INTRODUCTION
Image registration is an inevitable problem arising in many remote sensing applications whenever two or more images of the same scene have to compared pixel by pixel. These applications include: multisource data fusion, change analysis, image mosaicking, and scene matching.APPROACH
A new feature-based approach, as shown in the flowchart, to automated image-to-image registration is proposed. The characteristic of this approach is that it combines moment invariant shape descriptors with improved chain code correlation to establish correspondences between the potentially matched regions detected from the two images. It is robust in that it overcomes the difficulties of control point correspondence by matching the features first in the feature space using the minimum distance of the combined criteria, and sequentially in the image space using the rule of root mean square error. In image segmentation, the performance of the Laplacian of Gaussian operator was improved by introducing a new algorithm, called Thin and Robust Zero-Crossing, for searching, sorting, and refining edge points. The centers of gravity were then extracted from the matched regions and used as control points. Transformation parameters were estimated based on the final matched control point pairs.RESULTS
- The performance of the proposed algorithm has been demonstrated by registering two multitemporal Landsat TM images taken in different years.
- Registration accuracy of one-third of a pixel has been achieved.
- The proposed automated algorithm outperforms manual registration by over half a pixel, on the average, in terms of the RMSE at the GCPs, as shown in the table below.
- The technique of automated image registration developed in this work is powerful and reliable in terms of its registration accuracy, computational efficiency, and degree of automation.
- The uniqueness of this approach is also its robustness since it overcomes the difficulties of control point correspondence in the process of image matching caused by the problem of feature inconsistency
Contact the Webmaster Contents Copyright © 2000 NC State University
