We propose a method for recovering handwriting from background printing
and from underlying pages. Photometric stereo
(PS) is used to recover
the shape of the page surface. Surface models and signal processing
techniques are used to enhance indentations made by the pen. When tested
with paper lit with raking light, PS was found to accurately estimate
facet slope within the required range. It was also found to be effective
at separating handwriting from background printing---provided that
the printing was not too dark. We found that matched filtering was
the most effective signal processing approach to recover indentations
from underlying pages, but that spectral subtraction was most effective
at reducing background variation. We conclude that the combination
of PS and signal processing is a powerful tool for document analysis
with both forensic and commercial applications.
Applications
Using Photometric Stereo (PS)
