Abstract: In recent years,Bayesian approaches using Gaussian mixture model as a patch prior has achieved great success in image restoration.Aiming at the shortcomings of fix components and mainly relying on external learning of these models,a new image prior model based on Dirichlet process mixture model is proposed.The model learns external generic priors from a set of external clean images and learns internal priors from a given degraded image.Due to the accumulative property of the statistics in the mo-del,the integration of internal and external priori is naturally achieved in image restoration.Through the add and merge of cluster components,the model complexity can be adaptively changed as the data increases or decreases,more interpretable and more compact models can be learned.In order to solve the variational posterior of all hidden variables,a scalable variational algorithm combining with batch update with birth and merge mechanisms is proposed.The new algorithm improves the traditional coordinate ascent algorithm which is relatively inefficient under large data sets and often falls into the local optima.The effectiveness of the proposed model is verified by image denoising and inpainting experiments where the proposed model has advantage both on objective quality assessments and on visual perception comparing to traditional methods.

Key words: Batch update, Dirichlet process mixture model, Image restoration, Prior learning, Variational inference