How to obtain the posterior probability of 2D distribution?

Reversible jump MCMC for variable numbers of parameters used for Dirichlet process (DP) modeling of number of clusters is unlikely to work in practice. Clustering is hard enough already as a combinatorial problem. One thing you can do is take a large number of clusters relative to the data and then you can approximate the DP model. It will still be a challenge to fit, especially if the data is high dimensional and heterogeneous.