Unwrapping large interferograms using the minimum cost flow algorithm

For many applications of SAR interferometry phase unwrapping is an essential processing step. However, though unwrapping errors usually result in large error contributions compared with thermal or decorrelation noise, algorithms for fast and reliable phase unwrapping are still under investigation. Phase unwrapping becomes difficult when the signal is affected by strong decorrelation noise (repeat pass interferometry) or the signals from the scattering objects are not sampled correctly (layover). At DLR a digital elevation model (DEM) processing system is currently developed for ERS Tandem data and Shuttle Radar Topography Mission (SRTM) data, that relies hardly on correctly unwrapped phase values. The weighted least squares approach used until recently seemed to give locally good results, but it turned out to produce slowly varying large scale errors. In 1996 Costantini (1996, 1998) proposed a new branch cut based phase unwrapping algorithm, that minimizes the total weighted length of branch cuts in the image (minimum cost flow). The present authors' experiences with a prototype implementation were promising with some restrictions. The giant memory requirements of the general purpose minimum cost flow code limited the usability to smaller scenes (2700/spl times/2700) even on a larger computer. In response to this problem the authors developed a new MCF implementation, optimized with respect to memory and speed for InSAR phase unwrapping. Another problem is to correctly guide the branch cuts through alpine or forested terrain. The authors present cost functions that improve the unwrapping result in such regions.