Digital restoration of historical documents by diversity techniques and statistical processing

Using spectral or spatial diversity associated with statistical processing has been proved useful to restore degraded texts in historical documents. By linear independent component analysis, we have been able to separate the main text from interfering patterns or hidden features in color or multispectral document images, and to cancel the showthrough-bleedthrough distortion from suitably registered graylevel recto-verso document images. By applying the same principles to RGB recto-verso images, we have now demonstrated that the recto and verso patterns can be separated as in the graylevel case, and their original colors can be reconstructed. Some examples from real documents will be shown.