Check-actuated control mechanisms – Control mechanism actuated by check – other than coin – which... – By pliant currency
Patent
1996-06-19
1999-11-30
Bartuska, F. J.
Check-actuated control mechanisms
Control mechanism actuated by check, other than coin , which...
By pliant currency
194317, G07D 700
Patent
active
059926007
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND OF THE INVENTION
This invention relates to a money validator for validating articles of value; particularly but not exclusively, to a money validator for validating bank notes.
Money validators are known which comprise sensors (e.g. a plurality of optical heads) for generating sensed signals in response to an article to be validated, and decision means (e.g. a microprocessor, microcomputer or LSI circuit) for decided, on the basis of the sensed signals, whether the article corresponds to a valid denomination or not. Generally, the validation process may consist of comparing the sensed signals (or values derived therefrom), with predetermined thresholds which define acceptable ranges or "windows" of values, within which signals from a valid article are assumed to lie and outside which signals from invalid articles are likely to lie.
It is known to vary the thresholds, and/or the ranges or windows, over time to take account of drift, either in the response of the sensors, or in the properties of a population of articles to be validated; see, for example, European patent 155126 and International application WO80/01963 (British Patent 2059129), or European published Application EP-A-0560023 and corresponding U.S. application Ser. No. 08/013,708.
This technology is found to be effective where there is a clear limit boundary between a genuine sample population and a false sample population. However, this is not always the case. Separating the genuine from the false population, in the measurement space defined by the sensor output signal axes, may require a complex, non-linear boundary.
A pattern recognition technique known in the general field of pattern recognition is the so called "neural network" technique. In neural network techniques, sensor output signals are supplied to a plurality of similar parallel processing units, the outputs of which comprise some function of their inputs. In practice, the "units" are usually not provided by separate hardware, but by a single processor executing sequential processing.
Neural networks generally operate in two phases. In a training phase, the functions applied by each unit are derived by an interative training process comprising presenting known genuine and false samples to the network. In the case of a "supervised" type of network, such as the back propagation or perceptron type, the outputs of the units are monitored, compared with a "correct" network output, and the difference or error between the actual network output and the correct output is propagated back to incrementally affect the functions applied by each unit to its inputs. Where two or more network layers are provided, the function is a non-linear (e.g. sigmoidal) function of the weighted sum of the inputs, which enables the network to discriminate between disjoint patterns, and allows complex non-linear pattern separation discriminants.
After the training phase is complete (and, typically, the training phase may require many hundreds of thousands of iterative adjustments to the network) the network is found to have adapted to the statistics of the population of true and false articles to be validated which was presented during the training phase, and can generally perform a non-linear (in measurement space) separation between true and false articles which are subsequently presented to the network.
Hitherto, however, neural networks have not been applied to banknote recognition. Indeed, there would be resistance to the use of neural network techniques in money validation, because it is difficult to accurately characterise the level of performance of the network since, due to the non-linear nature of the signal processing applied by the network, it is not always clear how effectively, or in what manner, the network is discriminating between true and false articles.
SUMMARY OF THE INVENTION
In one aspect, the present invention provides a money validator comprising a classifier which classifies articles as one of a number of predetermined types of articles of value utilising classification dat
REFERENCES:
patent: 4556140 (1985-12-01), Okada
patent: 4578770 (1986-03-01), Mitani
patent: 5052043 (1991-09-01), Gaborski
patent: 5060278 (1991-10-01), Fukumizu
patent: 5067604 (1991-11-01), Metcalf
patent: 5201395 (1993-04-01), Takizawa et al.
patent: 5247584 (1993-09-01), Krogmann
International Search Report in International Application No. PCT/IB94/00375.
Examiner's Report to the Comptroller under Section 17 (The Search Report) in GB application No. 9324539.7.
"High Speed Conveyed Bill Money Recognition with Neural Network", Fumiaki Takeda, Sigeru Omatu, Takashi Inoue, and Salzo Onami; Robotics Mechatronics and Manufacturing Systems, Elsevier Science Publishers B.V. (North-Holland); 1993 IMACS.
"Bill Money Recognition by a Small Size Neural Network", Fumiaki Takeda, Sigeru Omatu, and Takashi Inoue; Bulletin of Faculty of Engineering, The University of Tokushima, No. 38, 1993.
Bartuska F. J.
Mars Incorporated
LandOfFree
Money validator does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Money validator, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Money validator will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1661176