Image analysis – Pattern recognition – On-line recognition of handwritten characters
Reexamination Certificate
1999-05-26
2002-02-19
Johns, Andre W. (Department: 2621)
Image analysis
Pattern recognition
On-line recognition of handwritten characters
C704S239000
Reexamination Certificate
active
06349148
ABSTRACT:
BACKGROUND OF THE INVENTION
The invention relates to a device for the verification of time-dependent, user-specific signals which includes
means for generating a set of feature vectors which serve to provide an approximative description of an input signal and are associated with selectable sampling intervals of the signal;
means for preparing a hidden Markov model (HMM) for the signal;
means for determining a first probability value which describes the probability of occurrence of the set of feature vectors, given the HMM, and
a threshold decider for comparing the first probability value with a threshold value and for deciding on the verification of the signal.
For the verification of time-dependent, user-specific signals, notably signatures or speech signals, it is checked whether an input signal indeed originates from a specific user or is a forgery. In this context the term “time dependency” is to be understood to mean that the signals are venerated by the user in a giver time interval, specific, different signal components being associated wit different instants within the time interval. Before verification can take place a signal model must be formed by means of one or more original signals; for this purpose use is made of so-called hidden Markov models (HMMs). The original signals used for forming the model are training signals for the so-called training of the HMM model. After completion of training, a signal can be verified by means of the device. To this end, a user identification, for example a user name or a number assigned to the user, is entered on the one hand and the user-specific signal on the other hand. The input signal is transformed into a set of feature vector. In order to form the vector components in the case of signatures, for example co-ordinates passed during the writing of the signature are evaluated and also the pressure exerted by an input stylus. Subsequently, there is formed a probability value which describes the probability of occurrence of the set of feature vectors for the HMM model assigned to the user with the user identification. The input signal is recognized as an original signal up to a selectable threshold value and beyond that as a forgery.
In devices of this kind, however, a problem is encountered in that an effective improvement of the threshold determination, and hence an effective improvement of the error rate, by increasing the number of original signals used for the training necessitates a disproportionally large number of additional original signals which often are not available prior to the putting into operation of the device.
SUMMARY OF THE INVENTION
Therefore, it is an object of the invention to improve the device of the kind set forth in such a manner that an improved threshold value determination and an enhanced error rate are achieved without using additional original signals for the verification.
This object is achieved in that the threshold value is dependent on an automatically determined, person-dependent second probability value which is formed by means of training signals used for training the HMM model and at least one additional validation signal which is not used for the training.
The subdivision of the group of original signals available prior to the putting into operation of the device into signals which are used exclusively for the training of the HMM model and at least one signal which is not used for the training of the HMM model but exclusively as a validation signal for improving the threshold value offers au effectively enhanced error rate for the verification by means of the device. The device automatically determines a person-dependent second probability value for each user separately, the person-dependent threshold value to be determined being dependent on said second probability value. Preferably, the second probability value is formed by forming an average value, notably the arithmetical mean value, of the first probability values which are formed upon input of the validation signals after the training of the HMM model. Thus, an average value of validation signal probability values is formed.
The threshold value is formed notably by the sum of the person-dependent second probability value and a user-independent constant, so that the error rate can be further improved. Generally speaking, the term probability value is to be understood to describe a value which is derived from a probability, notably the original value of the probability, or a logarithmic value of the probability.
The invention is preferably used for on-line verification, but is also suitable for off-line verification. The user-specific signals are, for example, signatures or speech signals.
Embodiments of the invention will be described in detail hereinafter with reference to the drawings. Therein:
REFERENCES:
patent: 5231381 (1993-07-01), Duwaer
patent: 5839103 (1998-11-01), Mammone et al.
patent: 5848388 (1998-12-01), Power et al.
patent: 5995927 (1999-11-01), Li
K. Fukunaga, “Introduction To Statistical Pattern Recognition”, 2NDEdition, Academic Press, New York, 1990, Chapter 10.1 and 10.2.
L.R. Rabiner And B.H. Juang: “Fundamentals Of Speech Recognition”, 1STEdition, Prentice Hall, 1993, Chapters 6.4. To 6.6.
L. Yang, B. Widjaja And R. Prasad, “Application Of Hidden Markov-Modela For Signature Verification”, Pattern Recognition 28, pp. 161-170.
Azarian Seyed
Johns Andre W.
Piotrowski Daniel J.
U.S. Philips Corporation
LandOfFree
Signal verification device does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Signal verification device, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Signal verification device will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2980154