Reconstructing missing regions in colour images using multichannel median models

This paper presents a method for reconstructing missing regions in colour images. A multichannel median model is proposed as the underlying image model and a statistical framework is employed to generate sampled realisations of the missing data. The nature of the model leads to a posterior expression for the missing data that does not involve an easy to manipulate multivariate probability distribution. Therefore, the problem of sampling is solved using a numerical approach. Results are included which show that this approach leads to excellent reconstructions.