Time-space filtering of multitemporal SAR images

Abstract

Most applications using multitemporal SAR images would benefit from speckle reduction if this could be done without destroying the spatial resolution or the temporal information. The paper proposes a time-space filtering method for multitemporal sets with uncorrelated speckle. For scenes with vegetal cover, the uncorrelation condition is easily fulfilled even for short acquisition time scales. To benefit from this property, the method combines two powerful tools: (i) a discrete cosine transform (DCT) applied on the time axis and (ii) a spatial adaptive filter applied on the resulting images which correspond to the different DCT coefficients. The results obtained on various sets of multitemporal radar images from satellites ERS1/2 show an important speckle reduction whereas initial radar reflectivity and spatial features are very well preserved.