Background

The Netherlands Forensic Institute (NFI), located in The Hague, is in search for a signature verification system that can be implemented in forensic casework and research.

The use of automatic signature verification tools can aid the forensic handwriting experts in drawing their conclusion about the authenticity of a questioned signature, but is not widely accepted nor implemented in most forensic laboratories. In our opinion, we need to bridge the gap between recent technology developments and the implementation of new automated tools in forensic casework. As a first step, the NFI wants to take the lead in comparing different signature verification algorithms systematically for the forensic community, with the objective to establish a benchmark on the performance of such methods.

The objective of this competition is to allow researchers and practitioners from academia and industries to compare their performance in signature verification on a new unpublished forensic-like data set. Genuine signatures and skilled forgeries were collected while writing on a paper attached to a digitizing tablet. The collected signature data are available in on- and offline format. Participants can choose to compete on the online data or offline data only, or one can choose to combine both data.

In forensics, online signature verification is not yet a common type of criminal casework for a forensic expert because questioned signatures and the collected reference (known) signatures are commonly supplied offline. However, it cannot be excluded that forensic experts will receive questioned online signatures in the near future. Furthermore, systems that combine both on- and offline information could also become of interest to the forensic experts, if such a system appears to be helpful in verifying an offline questioned signature to known signatures collected on a paper attached to a digitizing tablet by the criminal investigation department. If one chooses to use the combined data formats in the competition, we allow participants to use the combined information for the reference and the questioned signature. However, a more forensic-like situation would be to compare a questioned offline signature vs. a combined on-offline reference sample. So, we encourage participants to test their systems on this scenario also.

This competition is not an official certification test because we cannot use real forensic data, but we attempted to collect our data in a way that is close to real forensic casework. Of course we realize that the performance of a system can vary significantly with how forgeries are provided. Therefore, we offer a very large number of skilled forgeries in the training phase of the competition than in the evaluation phase. The training and evaluation phase consist of highly comparable data, collected under similar conditions and digitized with similar image scanner and WACOM Intuos2 tablets, yet the data were collected in different laboratories in different years.

We hope that, with this competition, researchers can identify areas where possible improvements to their algorithms could be made. Next to that, we have the intent to publish the results and to present the results of the competition at international forensic conferences and at other forensic laboratories. When the committee publishes an article, we will describe the evaluation data in general terms, which can be referred to in future publications. This publication will also include anonymous results from participators. Publications including a description of the data distributed in this competition should be sent in for a four week review by the NFI. For correct wording of the description of the data, we will publish a sample text on the website that will describe the database of this competition. The evaluation dataset will be made available to the competitors after the competition.

We invite all researchers and developers in the field of signature verification to register and participate in the ICDAR 2009 signature verification competition for off- and online skilled forgery data.