Ink classification and visualisation of historical manuscripts: Application of hyperspectral imaging

Overwritten characters and mixed inks can make documents hard to read. In this paper, we present and discuss research aimed at enhancement of a historical manuscript by separating text written using two different inks in two directions. We have used hyperspectral imaging to acquire images of the manuscript in the visible and near-infrared spectral region. The obtained data represents the spectral reflectance at every point of the manuscript surface. Visualisation of the manuscript under different illuminant has been simulated and demonstrated the issues associated with conventional digitisation techniques using 3-channel imaging. Spectral classification has been performed using two classification algorithms; spectral angle mapper and spectral information divergence. We have thus demonstrated the usefulness of spectral imaging to separate the inks.