Our model for tree crown extraction 2/2

Proposition kernel

As explained before, we use a Reversible Jump MCMC algorithm in order to sample the process. The global proposition kernel contains a uniform birth and death kernel ($Q_{BD}$), a translation kernel ($Q_{T}$), a dilation kernel ($Q_{D}$), a rotation kernel ($Q_{R}$), a split and merge kernel ($Q_{SM}$), and a birth and death in a neighbourhood kernel ($Q_{BDN}$). Figure (4) shows some of these moves.

Figure 4: Lefthandside : non jumping kernels, i.e. moves that do not affect the number of objects of the configuration. Righthandside : split and merge kernel, the grey shapes can be seen as the trees we want to extract.