In order to map damaged areas, a comparison (image differencing)
between pre- and post-disaster conditions using different polarimetric
features mentioned above has been conducted as a first step. After generating
a M×N co-registered difference image, XD= {X(i,j), 1≤i≤M,
1≤j≤N}, we aimed to classifying the area into {ωn,ωc} by thresholding
the value of To, where ωn and ωc are the classes associated with
unchanged and changed pixels, respectively. Most studies of image
thresholding tend to be based on parametric approaches that assume
a predefined statistical model for approximating the class distributions.