Automatic ink mismatch detection for forensic document analysis

A key aspect of handwritten document examination is to investigate whether some portion of the text was modified, altered or forged with a different pen. This paper demonstrates the use of hyperspectral imaging for ink mismatch detection in a handwritten note. We propose a novel joint sparse band selection technique that selects informative bands from hyperspectral images for accurate ink mismatch detection. We have developed an end-to-end camera-based hyperspectral document imaging system and collected a database of handwritten notes which has been made publicly available. Algorithmic solutions are presented to handle specific challenges in camera-based hyperspectral document imaging. Extensive experiments show that the proposed band selection method selects the most informative bands and improves average accuracy up to 15%, compared to using all bands.