Statistical two-dimensional edge linear prediction with fast algorithm

摘要

Two-dimensional linear prediction algorithms the past 30 years have predicted to a single point [1]. A more proper approach that fully utilizes the two-dimensional statistical information in the 2-D autocorrelation is to make a 2-D linear prediction based on an edge prediction rather than a single point prediction. This paper introduces this 2-D linear prediction edge approach and integrates it into a computationally fast 2-D minimum variance spectral technique. The improvement in edge-based 2-D linear prediction is illustrated with an actual data source of complex synthetic aperture radar phase history data. It is shown that the new edge approach creates a fully 2-D white residual from the 2-D linear prediction algorithm.