Unsupervised Bayesian classification of remote sensing images usinga pyramidal structure and the SEM algorithm

摘要

The pyramidal image decomposition consists of a succession offiltering and undersampling giving little smoothed images. Details areremoved allowing a better classification at each level of the pyramid.The SEM, a stochastic version of the EM algorithm, has been chosen forthat. At some level, pixels are split into two categories: those, whichcome from parents already classified and those, which have not beenpreviously classified. The classification is done on these last. Then,according to an homogeneous test, some of these pixels are considered aswell classified and the other will be classified at the next levelagain. Going from the top to the bottom of the pyramid leads to thewhole image classification. This procedure has been studied for variousconditions of filter type, homogeneous criterion and applied tosimulated and remote sensing images