Image restoration and segmentation using region-based latent variables: Bayesian inference based on variational method

摘要

To represent edges in image processing based on Bayesian inference, it is very effective to introduce latent variables. In this paper, we derive a deterministic algorithm that restores and segments an image using region-based latent variables and variational inference. This algorithm estimates two hyperparameters as well as infers the original image and the latent variables. In addition, the algorithm carries out model selection by minimizing the variational free energy. Through experiments using an artificial image generated by the heat bath method and natural images degraded by Gaussian noises, the effectiveness and limitations of the derived algorithm are shown.