Bayesian model selection for plane reconstruction

A key problem when fitting a model to data consists in the appropriate choice of the complexity of the model. On one side, the more complex model (that is, a model with more parameters) will usually follow the data more closely. On the other hand, the more complex model is likely to over fit the data. Several solutions to this problem have been proposed in the literature. Basically, the complexity of the model has to affect the posterior probability of the model given the data. This paper describes model selection using a Bayesian approach. It will turn out, that a term, punishing excessive model complexity, arises automatically from the calculations. To complete this report, the application of the results to building roof reconstruction is briefly shown.