Quality improvement of multispectral images for ancient documents analysis

Multispectral imaging is widely used for the analysis of ancient documents, like manuscripts or printed books, affected over time by degradations. Document digitization is performed with a monochrome sensor (CCD or CMOS) and an optical filter for each spectral band (infrared, visible, ultraviolet). This allows to capture additional information with respect to common RGB imaging, revealing details invisible to human eye. The use of optical filters causes geometrical changes and channel-dependent blurring, due to different refraction indices and manual focus setting. Moreover, document manipulations cause alterations among the channel images. Then, if the purpose of the study is the virtual restoration of the document and not only its interpretation, elaboration of the multispectral images is required. In this paper we will present a methodology, tested on individual solutions, to preliminarily register the images in order to correct geometrical misalignments, and apply deblurring techniques to improve image quality for further document analysis, where sharper images are needed. Deblurring is performed with a multichannel approach, adapted to the case of multispectral acquisition. Statistical techniques of decorrelation are applied to improve the legibility of documents or to restore degraded features, extract individual context parts of the document, separate patterns as the main text from the background, or attenuate interferences due to the seeping of ink from verso-to-recto.