Hyperspectral Imaging for Ink Mismatch Detection

EasyChair Preprint 3883

Abstract

 The discovery of ink mismatch provides important clues to pre-writing examiners to indicate whether a particular manuscript is written by a specific pen, or if a particular part of a note (e.g., signature) is written on a different ink compared to any other ink. In this paper, we show that the hyperspectral image (HSI) of the handwritten notes differs between visually similar inks. For this purpose, we have created the first hyperspectral data domain for a handwritten image in various blue and black inks, containing 33 visual reference bands. In unsupervised clustering technique, visual responses of inks fall into different groups to allow the separation of two different inks from the text of the questions. The same method when used in the RGB scan of these outputs fails to accurately distinguish the ink because it is very difficult to separate the ink from the optical range. HSI overcomes RGB deficiencies and allows better discrimination between inks.

Keyphrases: Hyperspectral images, K-means clustering, ink mismatch detection.

@booklet{EasyChair:3883,
  author    = {Aqsa Khan and Sara Sheikh and Nazish Iqbal},
  title     = {Hyperspectral Imaging for Ink Mismatch Detection},
  howpublished = {EasyChair Preprint 3883},
  year      = {EasyChair, 2020}}