A preliminary study on the forensic application of Near Infrared Hyperspectral Imaging (HSI-NIR) for indented writing recovery

Indented writing is the impression from the writing that has been transferred to the sheets of paper below the one where it was properly handwritten and when recovered, can reveal important information to criminal investigation, such as determining illegal financial transaction values or identity of criminals.
From the non-destructive techniques available, oblique lighting/photography and electrostatic detection are the most used by forensic document examiners. However, both techniques are unable to recover small indentations occurring more than three sheets down and require a skilled and experienced operator. Additionally, the electrostatic detection is affected by the type and quality of the paper, humidity and the ability to charge the surface of the document that retain the toner particles.
This study is a preliminary investigation of the potential of hyperspectral imaging in the near infrared range (HSI-NIR) combined with texture analysis of the images, as a rapid, objective and alternative method to the recovery of indented writings.
Samples consisted of two capital letters (A and C) and two numbers (4 and 6) that were written by the same person on the first of six stacked regular white office paper sheets. The writing was performed under normal and heavy pressures using four different types of black pens (ballpoint, rollerball, gel and hydrographic).The five sheets below the original writing were analyzed using the HSI-NIR camera SisuCHEMA SWIR from Specim in the spectral range of 928-2,524 nm. The images were obtained with two types of lenses (10 and 50 mm of spatial field), resulting in pixel sizes of 30x30 µm and 150x150 µm, respectively.
The best preprocessing evaluated consisted in changing the spectral range to 1,080-2,487 nm in order to include only the informative parts and correcting the light-scattering using Standard Normal Variate. Some other preprocessing techniques such as mean center and Savitzky-Golay smoothing, as well as the raw spectra, were also tested, but they did not show satisfactory results.
Principal Component Analysis (PCA) was initially explored and the image scores of the first component clearly revealed the indented writings of the first page made under hard pressure for three types of pens (ballpoint, rollerball and gel). When performing PCA considering not only the intensity of the nominal pixel but also the intensity of the eight surrounding pixels, the quality of the score images recovered improved. Future investigations require the use of other texture analysis techniques.