Image analysis – Applications – Personnel identification
Reexamination Certificate
1999-08-06
2002-03-12
Mehta, Bhavesh (Department: 2621)
Image analysis
Applications
Personnel identification
C382S209000
Reexamination Certificate
active
06356650
ABSTRACT:
BACKGROUND OF THE INVENTION
The invention is directed to a method for the adaptation of a reference data set on the basis of at least one input data set, using a computer.
Known methods for the verification of the subscriber, for example by inputting a password or with a chip card, are based on the fact that the subscriber has knowledge (password) known only to him or that only he has the corresponding medium (chip card) for granting access.
As an alternative thereto, biometry makes use of a physiological or of a behavior-typical mannerism of the subscriber in the automatic identity verification or in general for authentification.
Physiological methods thereby make use of a human mannerism that normally does not change or changes only slightly. Corresponding features exhibit the advantages that they cannot be stolen and can only be copied with great effort.
A signature verification is a biometric method. A signature input into a computer is thereby processed independently of the textural content, with the goal of either confirming or denying the authenticity of the author.
German Patent 195 11 470.1-53 discloses that a signature be obtained in electronic form using an electromagnetic pad. Such an electronic signature comprises a plurality of characterizing quantities, for example coordinate information, pressure and speed at respectively discrete sampling times. A verification of an input data set is based on a comparison with a reference data set. In the case of a signature verification, what is involved in the reference data set is an electronic signature, referred to below as “reference signatures”; the input data set is a currently requested signature that, for example, is input with an electronic pad.
It is well known that hand-written signatures of the same person are not absolutely identical to one another. Moreover, critical features of the signature of a person can change over time.
The reference by T. K. Worthington, T. J. Chainer, J. D. Williford, S. C. Gunderen (IBM Dynamic Signature Verification, Computer Society, IFIP 1985, pages 129-154); and discloses a method for reference data adaption. However, there is a risk that a forgery can occur.
SUMMARY OF THE INVENTION
The object of the invention is comprised in automatically adapting reference data that is subject to modification over time without being subject to forgery.
The invention specifies a method for the adaption of a reference data set on the basis of at least one input data set, using a computer. To that end, an evaluation criterion that characterizes a coincidence with the one reference data set is thereby first defined for the at least one input data set. When the evaluation criterion lies within a predetermined, first range, the input data set is entered into a buffer memory and a new reference data set is determined. When the evaluation criterion lies within a predetermined, second range, then a next reference data set is determined from the data sets of the buffer memory. When the next reference data set is “better” than the “old” reference data set, then the next reference data set is employed instead of the “old” reference data set. When, by contrast, the next reference data set is “poorer” than the “old” reference data set, then the next reference data set is discarded. When the evaluation criterion lies within a third range, then the reference data set remains unmodified.
A first data set is thereby “better” than a second data set when a distance of the first data set from the reference data set defined by an evaluation criterion is shorter (i.e. coincides better with the reference data set) than the distance of the second data set from the reference data set. Analogous thereto, the second data set is then “poorer” than the first data set.
The reference data set is preferably determined in that prescribable characteristic quantities of a prescribable plurality of data sets are averaged. As known from German Patent 195 11 470.1-53, there is thereby a data set (vector) at each sampling time that contains the characteristic quantities for this sampling time. Taking all or a part of these characteristic quantities into consideration, a reference data set is determined by averaging the values.
Alternatively thereto, the reference data set can be selected from a set of original data sets, i.e. data sets wherein it is certain that they derive from the authorized subscriber, this reference data set representing the best description of the original data sets in the feature space defined by the characteristic quantities.
Without limitation, these two possibilities are merely two alternatives for reference data set determination. Many other possibilities are conceivable, these likewise being capable of being employed in the invention.
One development of the buffer memory is a ring buffer that contains a prescribable plurality of data sets. A ring buffer is characterized in that the oldest data set therein is erased as soon as a new data set is added. This, of course, applies given the condition that the ring buffer is full since the new data set is otherwise simply added.
A next development of the invention is comprised in verifying the input data set as an original data set when the evaluation criterion lies within the first range or the second range. When the evaluation criterion for the input data set lies outside the first or outside the second range, then a verification of the input data set is to be negated; the input data set is considered to be a forgery.
Another development is comprised therein that original data sets are written according to a probability distribution. Such a probability distribution can be a normal distribution with an anticipated value and with a standard deviation.
Within the framework of an additional development, further, the reference data set is determined by the anticipation value of the original data sets.
Finally, a subdivision of the individual ranges can be determined by the following notation:
First range:
BEW &egr; [0; &mgr; + &agr;
1
&sgr;]
Second range:
BEW &egr; [&mgr; + &agr;
1
· &sgr;; &mgr; + &agr;
2
· &sgr;],
Third range:
BEW &egr; [&mgr; + &agr;
2
· &sgr;; ∞],
whereby
BEW indicates the evaluation criterion,
&mgr; indicates the anticipation value,
&sgr; indicates the standard deviation,
&agr;
1
indicates a prescribable parameter with which an upper threshold of the first range is identified;
&agr;
2
indicates a prescribable parameter with which an upper threshold of the second range is defined.
It is also possible to additionally subdivide the value range of the evaluation criterion into a fourth range that lies between the second and the third range and is defined with prescribable thresholds. Original data sets that are recognized as such without the reference data set being modified can lie in this range.
Within the scope of an application of the invention, the data sets can be signatures that are electronically registered and stored and that comprise a prescribable set of characteristic quantities (also see the introduction to the specification).
REFERENCES:
patent: 4724542 (1988-02-01), Williford
patent: 5052043 (1991-09-01), Gaborski
patent: 5479531 (1995-12-01), Webster
patent: 5559895 (1996-09-01), Lee et al.
patent: 5680470 (1997-10-01), Moussa et al.
patent: 5802200 (1998-09-01), Wirtz
patent: 5930380 (1999-07-01), Kashi et al.
patent: 29 45 377 (1980-05-01), None
patent: 195 11 470 (1996-10-01), None
patent: 2 104 698 (1983-03-01), None
Anthony, N. J. et al, Supervised Adaptation for Signature Verification System, IBM Technical Disclosure Bulletin, vol. 21, No. 1, (1978), pp. 424-425.
T. K. Worthington, et al, IBM Dynamic Signature Verifcation, Computer Security, (1985) pp. 129-154.
Mehta Bhavesh
Siemens AG
LandOfFree
Method for computer-adaptation of a reference data set on... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for computer-adaptation of a reference data set on..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for computer-adaptation of a reference data set on... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2861429