Hierarchical Bayesian segmentation of signals corrupted by multiplicative noise

The paper addresses the important problem of signal segmentation, when signals are corrupted by multiplicative noise. A hierarchical Bayesian analysis is proposed to estimate the change-point locations and amplitudes. However, closed form expressions of the change-point parameter estimators are difficult to obtain. The proposed methodology draws samples distributed according to the change-point parameter posteriors by a metropolis-within-Gibbs algorithm. The main advantage of the algorithm is that it allows joint estimation of the parameters and hyperparameters of the hierarchical model.