Application of log-cumulants to change detection on multi-temporal SAR images

Satellite SAR images acquired on repeated orbits became a useful source of information to monitor changes in many areas where optical data are rarely available. In this paper, two different approaches which take the specificity of speckle distributions into account are proposed to detect changing areas. The first one consist in detecting changes of the mean radiometry by applying in a temporal direction a conventional "edge detector". The second one consists in detecting temporally heterogeneous areas by measuring 3-dimensional texture parameters using second kind statistics. The results obtained with these two approaches and their complementarity are illustrated on a 7 image time series acquired by satellites ERS.