DD-SIMCA one-class modeling of blue ballpoint pen inks using hyperspectral imaging: forensic analysis of questioned documents

Alteration and addition of parts of numbers or text on documents is a common type of fraud that forensic document examiners often encounter in their casework. This type of fraud is typically committed to deceive or mislead others, and it can occur on a wide range of documents, from financial records and contracts to wills and deeds. Hyperspectral imaging (HSI) is a powerful tool for forensic analysis of document forgery [1-3]. The aim of this study is to investigate the issue of text addition and the order of crossed ink lines, which are commonly encountered in questioned documents. Twelve blue ink pens from different brands and models were employed to prepare the samples on A4 paper in two different ways: (i) a vertical line of 5 mm in length was drawn using each pen. (ii) A horizontal line was drawn with pen 1 and vertical lines were drawn with all 12 pens onto this line (pen1 under 12 pens). Out of the 144 possible pairwise comparisons between the 12 pens, 7 pair were selected for further analysis based on the criterion of the angle between the mean vectors of the pairs. Reflectance spectra were acquired using a linear scanning hyperspectral camera (HSI-Vis-12bit), manufactured by Parto Afzar Sanat corporation, between 400 - 950 nm with 2 nm step resolution. HSI data was performed on the samples at 5 different times, 1, 2, 30, 40, and 80 days after the lines were written. The image resolution obtained was 272x135 pixels and 536 wavelengths, which represents three-dimensional data. Three regions of each of the vertical lines were manually selected as regions of interest, resulting in a total of 168 pixels for one pen. The preprocessing methods used were conversion to pseudo-absorbance (log (-1/R)), smoothing using the Savitsky-Golay filter with a second-order polynomial and 20 points window, and Standard Normal Variate (SNV method. The problem at hand is formulated as an authentication problem, and a one-class classifier technique known as data-driven soft independent modeling of class analogy (DD-SIMCA) is employed to classify pairs of blue ballpoint pen ink samples. Classification results from DD-SIMCA yielded 97-100% sensitivity for the training set and 39–100% specificity for different test samples. The results indicate that HSI-Vis combined with DD-SIMCA is suitable for authentication of blue ballpoint pen ink on paper for questioned document examination except when the inks are very similar to each other.