Development of automated image co-registration techniques: Part II - multisensor imagery

摘要

This is the second in a series of PNNL Multispectral Imagery (ST474D) reports on automated co-registration and rectification of multisensor imagery. In the first report, a semi-automated registration procedure was introduced based on methods proposed by Chen and Lee which emphasized registration of same sensor imagery. The Chen and Lee approach is outlined in Figure 1, and is described in detail in the first report. PNNL made several enhancements to the Chen and Lee approach; these modifications are outlined in Figure 2 and are also described in detail in the first report. The PNNL enhancements to the Chen and Lee approach introduced in the first phase have been named Multisensor Image Registration Automation (MIRA). These improvements increased computational efficiency and offered additional algorithms for coarse matching of disparate image types. In the MIRA approach, one set of optimum GCP locations are determined based on a Delaunay triangulation technique using an initial set of GCPs provided by the user, rather than repeating this step for each added control point as is proposed by Chen and Lee. The Chen and Lee approach uses an adjacent pixel difference algorithm for coarse matching patches of the reference image with the source image, while the MIRA approach adds other algorithms. Also the MIRA approach checks to determine if the a newly determined GCP fits the existing warping equation.