Bayesian multichannel blind deconvolution for ancient document analysis and restoration

In the analysis and restoration of the content of ancient degraded documents, the main issue is often to separately extract and enhance the various layers of information overlapped in the document itself. We model multisensor images of a document as convolutive mixtures of the interfering patterns, and adopt a Bayesian estimation approach which exploits Gibbs priors, accounting also for well-behaved edges in the ideal images. We show applications to the removal of the bleed-through/show-through effects, and to the recovery of the original color of faded images. This latter application can be of interest in other cultural heritage contexts, such as the restoration of old photos and videos.