Automatic Co-Registration of Space-Based Sensors for Precision Change Detection and Analysis

A variety of techniques were developed at JPL to assure sub-pixel co-registration of scenes and orthorectification of satellite imagery to other georeferenced information to permit precise change detection and analysis of low (e.g. 1-4km weather satellite) and moderate (e.g. 30m Landsat) resolution space sensors. The methodology employs the additive composition of all pertinent dependent and independent parameters contributing to image-to-image tiepoint misregistration within il satellite scene. Mapping and orthorectification (correction for elevation effects) of satellite imagery defies exact projective solutions because the data are not obtained from a single point (like a camera) but as a continuous process from the orbital path. Standard image processing techniques can apply approximate solutions with sufficient accuracy, but some advances in the state-of-the-art had to be made for precision change-detection and time-series applications. The basic technique first involves correlation and warping of raw satellite data to an orthorectified Landsat data base to give an approximate mapping. Then digital elevation models are used to correct perspective shifts due to height and view-angle. The image processing approach requires from three (e.g. geosynchronous weather satellite imagery) to six (e.g. polar weather satellite imagery) sequential processing steps that warp the dataset by resampling pixel values. To avoid degradation of the data by multiple resampling, each warp is represented by an ultrafine grid of tiepoints. For successive warps, the grids can be composed mathematically into a single grid such that only one re-sampling occurs. Ultra-fine grids can currently be up to 1000 x 1000, or one million points.