Kernel smoothing classification of multiattribute data in the belief function framework: Application to multichannel image segmentation

Bayesian approaches turn out to be inefficient when decision making involves many uncertain, imprecise or unreliable sources of information. The same problem occurs in multiattribute data classification where each attribute can be perceived as a source of information. The theory of belief functions has been extensively used to overcome this issue. Moreover, Markov field approaches involving this theory have shown their interest in image modeling and processing. The aim of this paper is to propose a new approach for multichannel (typically remote sensing) image segmentation through hidden Markov fields, which better manage non Gaussian noise forms, by adopting Weighted Parzen-Rosenblatt Dempster-Shafer likelihood model. More explicitly, observations are modeled through belief functions constructed through a kernel smoothing- based scheme rather than using plain Gaussian densities as in the typical hidden Markov fields. The interest of the proposed approach is shown through experiments conducted on sampled and real multichannel remote sensing images.