Voice and signature to know who it is
2006/09/01 Kortabitarte Egiguren, Irati - Elhuyar Zientzia Iturria: Elhuyar aldizkaria
Regardless of its biometric feature, the system must have an appropriate sensor to read the biometric data. But equally important is to have a tool of comparison between the measured and the saved. That is, to have an adequate database. To repeat the sessions and compare the results of several algorithms it is necessary to use a rich database. It is also important for sensors to be able to detect counterfeiting efforts.
Faced with the need for a biometric database, the Department of Electronics and Telecommunications of the Technical School of Engineering of Bilbao began a few years ago the development of a database that collected the biometric characteristics of hundreds of people, among many others in Spain
with the university. Currently, to complete this database, we are working especially on the study of voice, signatures and writing.
As for the voice, the usual techniques generally use segmental characteristics to differentiate people. One of them is the so-called voice timbre, which can be improved despite good results. Precisely, UPV researchers want to improve this system by measuring the rhythm of voice and intonation, among others. Your intention is to integrate all these features into the database. In fact, it has been proven that when several biometric systems are combined or some parameters of the same biometric characteristic are added, the average error is always less than that of each independent system.
Signature on examination
In addition to the voice, they want to include the signature in this database. The automatic knowledge of the signatures can be divided into two large areas according to the form of data collection: online and offline signatures.
In offline signature or writing a document from the past is taken and scanned for image processing. All signature features will be derived from their spatial appearance. Therefore, counterfeiting is simpler, since the counterfeit only has to imitate the look of the firm.
In the case of online signature or writing, however, they not only analyze the spatial aspect of the firm, but also their dynamic information. They use a digitization table and a digitization pencil that collect at all times the trajectory of the pencil, the pressure or force that is exerted when writing, the inclination of the pencil, etc. When accessing the data on-line, the signatory must be present, since the movements of the pencil are measured as they are signed or written.
But how does the database develop? Each user of the signature database must make their own signature and imitate the signature of other users to somehow form the system.
Fewer errors
A safe biometric recognition system would not have to make any bad acceptance or denial, but in most cases it does. The error in online knowledge of signatures is approximately 4%. That is, 4% of real signatures and so many accepted counterfeits are discarded. As for offline knowledge, this figure is quite higher (20%).
In the case of the voice there has been a relatively small error. However, they are provisional results. In fact, different voice parameters are currently being combined for more accurate results.
