We have analyzed the problems in the visible and far-infrared battlefield image registration. We adopted the contour-based approach, which permits to incorporate the powerful computer vision approaches for edge detection, edge saliency and curve matching. We have implemented multi-scale hierarchical edges detection and edge focusing and salience measure for extracting horizon lines. To deal with the feature inconsistency problems we used the transformation space searching methods with the partial Hausdorff distance merit, and demonstrated free-form curve matching using 1) Image quadtree partition technique to the local Hausdorff distance matching, that dramatically reduces the size of the search space into that of the search for translations which minimize the Hausdorff distance between corresponding sub images, 2) Adaptive partial Hausdorff distance hill climbing, which is much faster than the search in the entire affine transformation space. 3) The iterative closest point algorithm using an efficient segment-list Voronoi transform and distance transform and again the mean square partial distance merit. All the three algorithms match the maps of curve features without looking for correspondence and show a large tolerance to outlier features. The condition for application of those three approaches is that the transformation should be known a priori and the initial match should be established as a start point for the optimization. Those conditions are fulfilled in our application. The key issue for the successful image registration is still the extraction of salient features from the real world images using local, regional and global information and appropriate salience measures, and an efficient curve matching against outliers.
展开▼
