Landsat Imagery Registration Using Correlation and Spatial Relationship Analysis

The historical archive of Landsat imagery allows performing multitemporal studies using a long series of images. For this purpose, images must be registered, that is its geometry must be made compatible. This task generally is performed using visible control points in the two images to estimate the transform parameters. The choice and visual detection of homologous points are arduous and laborious tasks. In this work an automatic method for the identification of homologous points is described, which uses the correlation and spatial analysis. As the digital correlation can identify false positives, that is, points with high correlation but that are far from the right position. This result needs a fine analyzing to select the right points, which is here performed analyzing the spatial relationship between the available points. In the experiments, an area with great diversity in the landscape was used, with intensive agriculture, dense urban occupation, roads and a river, allowing choosing and studying different control points. The results show that the process can be automated, but that the success of the process depends on the quality, in terms of contrast, of the used regions in the correlation

Keywords:
Image correlation; control points; geometric correction