Writer identification based on handwriting still is a labour-intensive job for human experts. The Trigraph project aims at automating this task to a certain degree in order to improve identification speed and accuracy. In this project three modes of writer identification are distinguished:
- mode 1: traditional, manual measurements (using computers);
- mode 2: automatically computed global measurements;
- mode 3: measurements on variatons in letter shapes, automatic to some degree.
Two research groups are participating: NICI in Nijmegen and the AI institute in Groningen. In our (Groningen) approach to the project, a writer identification system is developed which is to be used by the NFI, the Dutch forensic institute. We aim to satisfy two goals. First, the system should require only little human intervention, requiring a high degree of robustness with respect to the input material. Second, its output should be such that it can be used in court. The latter implies reliable identification performance, a sound statistical base and a cognitive presentation of the identification result. In order to reach these goals, research will be done on the improvement of all of the system's building blocks: preprocessing, feature extraction, classification and classifier combination. The project will build upon the work done by Lambert Schomaker and Marius Bulacu. The system will have two key aspects:
- Image-based: Feaures considered are on the level of picture elements (pixels). As such, writer identification of mode 2 and 3 will be performed. This will also bridge the current gap between Groningen's and NICI's approach to the mode 3 identification.
- Knowledge-based: Prior knowledge is used, such as statistical information from formerly treated documents and information entered by the user, such as writer nationality and age.
|
(part of) a questioned document |
|
suspect 1 |
suspect 2 |
suspect 3 |
suspect 4 |
suspect 5 |
suspect 6 |
|
