We develop a Bayesian nonparametric model for reconstructing magnetic resonance images (MRIs) from highly undersampled (k ) -space data. We perform dictionary learning as part of the image reconstruction process. To this end, we use the beta process as a nonparametric dictionary le