Image-to-image registration is the translation and rotation alignment step by
positioning two images that have from the same geographic area with respect to first
image to one another on top of one another (Jensen, 2005). We simply want to use these
two images to detect any changes that occurred on between two different dates. Since
both pre-and post-tsunami Landsat images are already assigned to the same projection
since we obtained, it is not necessary to again rectify post-tsunami Landsat image to the
topographic map again. Therefore, registering the post-tsunami image to the registered
pre-tsunami image should be acceptable. The post-tsunami Landsat image is registered
to the pre-tsunami image. Then, Nearest Neighbor technique is used for pixel
resampling. Nearest Neighbor interpolation better maintains original reflectance values
thus it is the only resampling method which should be performed before image
classification (Bottomley, 1998).
The goal of image-to-image registration is to make sure two images align
perfectly along the edges. Images of two different dates are registered to each other by
matching specific coordinate points. In this study, images are assigned to the same
projection and datum, but each single pixel does not necessarily match the corresponding
pixel in the other image perfectly. In our case, this problem can be caused by the
different spatial resolution of both images so the shifting technique provided in ArcGIS
helps to get exact alignment. The image of 2002 is then resampled to 25 meters to
coincide with the 2004 imagery. Since Landsat images of path 130/ row 54 have derived
35
from sources covering an area larger than our study area, subset cropping technique is
applied to narrow down the area. Subset cropping is a typical step of feature selection
helping to minimize computer storage and processing time. After the image preprocessing,
Landsat images from both dates are ready for image classification