Towards 6 m TanDEM-X DEM: Non-local Methods for InSAR Filtering

Abstract

The major goal of the TanDEM-X mission is to provide a high quality global digital elevation model (DEM) with post spacing of 12 meters. This study investigates intelligent InSAR filters for increasing the DEM resolution to 6 m while maintaining the accuracy. Due to the special nature of interferometric phase statistics the interferometric phase error of full resolution SAR interferogram pixels is usually higher than acceptable. Therefore several neighboring pixels are usually averaged in order to reduce the phase noise. Obviously this averaging leads to reduced resolution and blurring of image features, especially in urban areas or areas with strong topographic relief. Recent publications and investigations demonstrated that more intelligent averaging like non-local means filters can reduce the phase error while well retaining structures. Rather than averaging pixels in a local neighborhood such as rectangular windows, directional windows, or spatially connected components, non-local means filters also consider pixels that are far apart and weigh them according to their image patch similarity. Because non-local methods can use more pixels for averaging and take care of amplitude and phase features in the patches, they give in general much more reliable estimation of the amplitude, interferometric phase and coherence. In this contribution, we present our further development of the non-local filters by taking into account patch symmetry properties, and explore their potential for more accurate height estimation and multi-baseline phase unwrapping in order to support the TanDEM-X mission. Furthermore, for operational SAR data processing, the robustness of the developed filters is of great important and will be extensively tested with TanDEM-X data. Results of non-local means filtering of TanDEM-X data will be compared to conventional filters.