Classification of blue pen ink using infrared spectroscopy and linear discriminant analysis

Abstract

Attenuated total reflectance (ATR) Fourier transform infrared (FTIR) spectroscopy associated to linear discriminant analysis (LDA) was employed to perform classification of blue pen ink according to types and brands, in a nondestructive way. To build a representative data set, blue pens of 3 types, namely ballpoint (5 brands), roller ball (2 brands) and gel (3 brands) were purchased from local dealers. Ten different pens, representing the best seller of each brand, were purchased, making a total of 100 pens. Circular areas were painted five times with each pen and spectra were taken in 2 different locations, using a Universal Attenuated Total Reflectance accessory (UATR), within the range of 4000 to 650cm−1. Three types of paper were employed: two brands of A4 sulfite paper (paper 1 and paper 2) and one recycled paper (paper 3). The genetic algorithm (GA), stepwise formulation (SW) and successive projections algorithm (SPA) were employed to select spectral variables employed in LDA. LDA models were built using the blue pen ink spectra obtained from paper 1. Three test sets were employed using the blue pen ink spectra obtained from papers 1, 2 and 3, in order to evaluate the influence of the paper on the predictions. The LDA models used to classify the pens according to their type (gel, rollerball and ballpoint) achieved a correct classification rate of 100% in the test set composed of blue pen ink spectra obtained from paper 1, using GA and SPA. Using SW, the rate achieved was 99.5%. For paper 2, SPA, GA and SW provided 100%, 97.3% and 93.8% of correct classification, respectively. For paper 3, SPA, GA and SW achieved a correct prediction rate of 100%, 100% and 94.9%, respectively. LDA models for classifications of pens according to their brand were 100% correct in their classification when the test set was composed of blue pen ink spectra obtained from papers 1 and 2. For the test set composed of blue pen ink spectra obtained from paper 3, LDA-SPA, LDA-GA and LDA-SW classified them correctly at 91.3%, 100% and 100%, respectively. The method developed was able to differentiate successfully all brands of pen used on each type of paper and could be a helpful tool for detection and confirmation of counterfeits in documents of legal importance.