Bertrand Haas

Engineering better voting systems

We consider here an election ballot as a document, a document that works as the carrier of a voters choice in an elections accounting system for determining a winning candidate. And we consider a voting system as a way to manage both the document and its flow in compliance with the requirements of the election. Trustworthy elections are the core of a democratic spirit and engineering a voting system with requirements of convenience, privacy, integrity and reliability lies at the core of trustworthy elections. We define here some clear requirements for a trustworthy voting system and analyze the most popular classes of voting systems according to some of these requirements. We draw conclusions on the engineering of better voting systems and show two efforts in this direction.

Sub-pixel analysis to support graphic security after scanning at low resolution

Whether in the domain of audio, video or finance, our world tends to become increasingly digital. However, for diverse reasons, the transition from analog to digital is often much extended in time, and proceeds by long steps (and sometimes never completes). One such step is the conversion of information on analog media to digital information. We focus in this paper on the conversion (scanning) of printed documents to digital images. Analog media have the advantage over digital channels that they can harbor much imperceptible information that can be used for fraud detection and forensic purposes. But this secondary information usually fails to be retrieved during the conversion step. This is particularly relevant since the Check-21 act (Check Clearing for the 21st Century act) became effective in 2004 and allows images of checks to be handled by banks as usual paper checks. We use here this situation of check scanning as our primary benchmark for graphic security features after scanning. We will first present a quick review of the most common graphic security features currently found on checks, with their specific purpose, qualities and disadvantages, and we demonstrate their poor survivability after scanning in the average scanning conditions expected from the Check-21 Act. We will then present a novel method of measurement of distances between and rotations of line elements in a scanned image: Based on an appropriate print model, we refine direct measurements to an accuracy beyond the size of a scanning pixel, so we can then determine expected distances, periodicity, sharpness and print quality of known characters, symbols and other graphic elements in a document image. Finally we will apply our method to fraud detection of documents after gray-scale scanning at 300dpi resolution. We show in particular that alterations on legitimate checks or copies of checks can be successfully detected by measuring with sub-pixel accuracy the irregularities inherently introduced by the illegitimate process.