A fission and fusion Markovian approach for multi-channel segmentation

Proposes a fission and fusion approach for segmenting polar scenes from the multi-channel satellite AVHRR. The classes considered are sea, clouds, continent without relief, with relief and icebergs. The main difficulty, in a textural sense is to discriminate relief from icebergs. The authors have five channels from AVHRR to develop a fusion approach. Nevertheless, the information provided from initial captors is mixed. A simple fusion approach leads to a complex and often intractable model. Thus they start out by a fission step, introducing new captors containing simplified information, given by textural parameters. They then fusion these new data by a stochastic relaxation technique. The optimized model is a generalized Potts model in a non-homogeneous external field. This external field is composed of piecewise linear potentials representing the histogram modes. This simple model is enough due to the redundancy and simplification of information performed by the fission step. The Potts model induces regularization properties during the fusion step. The optimization is achieved by a simulated annealing procedure, letting temperature decrease geometrically. Results demonstrate the accuracy of the method. The fusion step is of paramount importance to simplify the estimation of parameters associated with data driven terms. The accuracy of the estimation method induces precise edges in the final result