Near and middle infrared hypespectral images to investigate forensic cases of suspected fraudulent bankchecks
Determining the fraudulent nature of a document is a common problem in Forensic Document Examination field, for example, alteration and addition of parts of text. Considering the high evidential value of some documents, the use of non-destructive techniques is necessary in order to assess the authenticity of the document without compromising its integrity. In this respect, Infrared spectroscopy has proven very useful in the identification and discrimination of inks in documents. Additionally, when combined with Chemometrics and Hyperspectral Imaging (HSI), information on the chemical composition and the distribution of inks can be obtained. This study examined suspected counterfeit bank checks. Blind tests were performed, in which an alleged forger had available 16 black pens (3 gel, 2 rollerball, 8 ballpoint and 2 hydrographic) from different brands and models to produce 10 samples (numbers in checks) either authentic or counterfeit (alterations or additions), reflecting real forensic casework. Another person, without any previous information, analyzed the samples. Near infrared hyperspectral images (HSI-NIR) were acquired using the HSI camera SisuChema from Specim. The images obtained consist of approximately 0.5 x 1.5 cm, pixel size of ~30 x 30 µm and at the range 928-2524 nm. Additionally, samples were analyzed by HSI at Middle Infrared (HSI-MIR) range (4000-750 cm-1) using the Spectrum 400 spectrometer (Perkin-Elmer) equipped with a Spotlight 400 microscope and the ATR imaging accessory. The size of these images was 100 x 100 µm and pixel size 6.25 x 6.25 µm. Finally, the images were corrected from undesirable effects related to noise and scattering radiation and analyzed with unsupervised chemometric tools such as Principal Component Analysis (PCA) and Projection Pursuit (PP). The best preprocessing technique obtained for NIR data set was Standard Normal Variate (SNV) with Savitzky-Golay smoothing filter (7 points window width and 2nd order polynomial); 1st derivative with Savitzky-Golay smoothing filter (9 points window and 2nd polynomial order) showed the best results for MIR data set regarding PP and PCA scores. Applying PCA and PP to the HSI-NIR, it was possible to solve 70% and 80% of the samples, respectively. In one case it was not possible to distinguish the inks from the paper (inconclusive) and the other was a false negative. For MIR data set, there were 2 false positives and one false negative. For the inconclusive case using HSI-NIR, the complementary use of HSI-MIR correctly determined that it had been forged/falsified. Using both HSI-NIR and HSI-MIR, just one conclusion was not correct (false negative), in which the registers were made by a gel and hydrographic pen. Therefore, HSI-NIR is a fitting technique for a rapid, non-destructive and initial analysis of suspected fraudulent documents and HSI-MIR should be used further as a complementary technique.