GB2039118A - A signature recognition system - Google Patents

Abstract

Apparatus for validating a complex analogue waveform, such as can be derived from a written signature or a spoken phrase, in which a number of different parameters are measured simultaneously and each measurement is compared with predetermined upper and lower limit reference values for that parameter. Acceptance or rejection of the waveform depends on how many of the parameters fall within their respective limit ranges.

Description

SPECIFICATION A signature recognition system This invention relates to apparatus for validating complex analogue waveforms and is particularly suited to the identification of individuals by utilising waveforms derived from a person's signature or speech.

There are many human activities in the modern world which demand rapid and reasonably conclusive proof of a person's identity. Probably the most common is proof of identity by way of signature on items such as bank cheques. Another sphere where rapid proof is required is in the ever increasing number of situations where it is deemed necessary to have some degree of security control, over access to specified premises for example. This can be done either by signature or voice, in the latter case a person would be required to speak his own name, for example.

However, in many cases it is possible for an unscrupulous person to imitate a spoken name or forge a signature. The purpose of the validation apparatus is to negate attempts at fraud. Perhaps the simplest case to consider is signature validation.

A validation operation is not to be confused with a recognition operation. Consider a bank's list of account holders (customers). The validation system is not required to identify one out of, say, a possible five million bank customers, but is faced only with the much simpler task of establishing with reasonable probability that the signatory is the genuine holder of an account already identified.

The normal method of validation, which involves comparing the visual pattern of the completed signature with an authentic signature held by the bank, while admirably suited to the highly-developed pattern recognition capability of the human eye and brain, suffers certain disadvantages. In particular, a potential forger, having obtained a sample of the victim's signature, is likely to have ample opportunity to practice and perfect a convincing replica, so that when the actual fraud takes place the forgery can be repeated with sufficient aplomb to avert suspicion. It would therefore be desirable that the validation process should depend upon features other than those staticised on paper for the thief to examine at leisure.

With constant repetition, a person's signature tends to become stylized in varying degree, some individuals producing a cryptogram bearing little relationship to their actual name, while even the most laboured performer will likely replicate certain characteristic loops and flourishes. As with other repetitive human behaviour, the detailed act of writing a signature sinks progressively below the conscious level until it becomes an automatic sequence of muscular reflexes. At this stage, the time taken to write the signature, the changing velocity of pen movement and the sequence of changes in the direction of pen movement are likely to be even more characteristic of the signatory than the final image as it appears upon the paper.Such features occur so rapidly that, even given the opportunity to watch the intended victim in the act of signing, a potential forger would find it virtually impossible to recognize the sequence, with even less likelihood of duplicating it.

Hence, a validation method can be based primarily on the comparison of waveforms derived from the velocity of pen movement across the paper, and pays no regard to the actual written pattern. This gives the added advantage that the complexity of optical pattern recognition electronics can be avoided entirely.

In the case of speech, whilst it is possible for one person to imitate, apparently accurately, another person's voice, nevertheless there are certain speech waveform characteristics that are peculiar to each person even when they imitate other people. These characteristics can be identified by analysis of samples and once they have been so identified they can be used as the basis of a validation system.

According to the present invention there is provided apparatus for validating a complex analogue waveform comprising a plurality of individual parameter measuring means each of which is arranged to measure a predetermined different parameter of the waveform, a same number of individual storage means each of which is arranged to hold upper and lower limit reference values for a different one of the parameters being measured, and a same number of comparator means each of which is arranged to compare the output of one measuring means with the stored reference values for the parameter being measured by that measuring means to provide an output indicative of whether or not the measured parameter value is within said limit values.

In a commercial application the apparatus would include a library of identity storage locations each of which comprises the number of individual storage means for one person's identity, and means for a person to enter a claimed identity prior to providing a complex waveform for validation, the apparatus including means for selecting the storage location corresponding to the claimed identity whereby the number of storage means in that location are coupled to the number of comparator means for the consequential validation operation.

Embodiments of the invention will now be described with reference to the accompanying drawings, in which: Figure 1 illustrates a basic apparatus for validating a complex waveform, Figure 2 illustrates part of an apparatus specifically designed to validate signatures, and Figures 3A-3C show different views of a signature sensing arrangement for use with the apparatus of Fig. 2.

In the arrangement shown in Fig. 1 a complex waveform to be validated is produced in a generator 10. Since the waveform to be validated is identifiable in advance of its generation this identity is entered, e.g. by means of a Keyboard, in a unit 11 before the waveform is produced. This identity is passed in the form of an access address to a store 1 2 which has a number of storage locations, one for each waveform identity. Each storage location is divided into a number of portions T1, .2... each of which holds upper and lower limit reference values for one parameter of that waveform. When a storage location is accessed these values are read out and held in a register 1 3. The complex waveform when produced is then subjected to a variety of different parameter measurements in a number of measuring means 14.For example, the following parameters may be determined: (a) Total duration of the waveform.

(b) The number of gaps, if any, in the waveform.

(c) The number of zero crossings in the waveform.

(d) The number of times the waveform exceeds a predetermined positive (or negative) threshold level.

These are only indicative of the sort of measurements that can be performed, others can readily be prescribed.

Each measurement is then compared in comparators 1 5 with the corresponding reference values held in register 1 3. The comparator logic can be arranged to deliver a decision output based on the number of measured parameters which fall within their respective reference limits. For example a validation may require all the parameters to be within limits, alternatively it may be set to require only a minimum number (n- xl out of n parameters to be within limits.

Reference has been made to particular applications for such validation apparatus, such as personal identification by way of speech or signature. Fig. 2 shows a block diagram of part of a signature validation apparatus. A signature sensing unit 20, corresponding to the generator 10 of Fig. 1, is coupled to the measuring means 11, corresponding to the measuring means 14 of Fig. 1. The signature sensing unit 20 is illustrated in more detail in Figs. 3A-3C which give respectively a plan view, side elevation and end elevation of the unit. The unit is housed in an open-topped rectangular box 31 covered by a rigid flat sheet 32 having a central rectangular aperture 33.Beneath the aperture there is a flat plate 34 mounted on a parallel motion suspension arrangement 35 to which is fitted a switch 36 the contacts of which are caused to open when slight downward movement of the plate 34 occurs under pressure from a writing stylus 37. The arrangement of the plate 34 and the sheet 32 is such that a supply of pressure sensitive paper 38 may pass beneath the sheet and over the plate. The box 31 contains two sensing coils 39a, 39b arranged vertically and parallel with the longer sides of the box.

The writing stylus 37 is a freely held magnetic pen.

When the stylus 37 is used to write a signature two separate electrical outputs are produced by the unit. The first output is derived from the switch 36 and is indicative of the number and duration of the pen contacts during the act of signing. The second output is derived from the two coils in which electrical currents are induced due to the movements of the magnetic pen. In a typical construction the two coils would be made of 600 turns each of 28 SWG enamelled wire, each having an inductance of 50 mH and a resistance of 600 ohms.When mounted close to the upper and lower edges of the signature area, relative to the normal signature writing direction, i.e. left to right, and connected such that their combined output voltage is a function of the vertical component of pen velocity they will respond only to such vertical movements of the pen in the signing area and will ignore horizontal pen movements. It will be appreciated that vertical pen movements in most signatures, and indeed in writing generally, are most frequent and usually attain higher velocities than the horizontal movements. A typical signature written on a unit such as that described produces a signal of a few millivolts amplitude in the frequency range 0.2Hz to 5Hz.

Returning now to Fig. 2, it can be seen how signals from the signature sensing unit 20 can be utilized.

Parameter A: Signature duration When the stylus 37 is pressed onto the plate 34 during the act of signing the contacts of switch 36 are opened. This action applies a voltage to a switch 21a which is then closed.

Closure of switch 21a allows a stream of pulses from an astable multivibrator 23 to increment a digital counter 22a. To be strictly accurate, this counter only determines the total time that the pen 37 is in contact with the plate 321 it does not take account of any short periods when the pen is briefly lifted from the plate during the process of writing the signature.

Parameter B: Number of pen contacts When the stylus touches the plate 34 a voltage is applied to close a switch 21b. This is turn applies a voltage to a counter 22b which is triggered each time switch 36 in the sensing unit is opened.

Determination of the remaining parameters is based on the output from the sensing coils 39a, 39b. A low-pass, fourth order Butterworth filter 25 is included in the sensing unit to remove any noise or interference above 5Hz, e.g. mains hum at 50Hz is attenuated by approximately 70dB. An amplifier 26 within the sensing unit raises the signal level to a value approaching one volt for transmission to the measuring circuits 11.

Parameters C, D, E and F: Threshold detec tions During the periods of pen contact a switch 24 is closed and the signal from the sensing coils is applied to a set of four high speed comparators 21c-21f, each of which is preset to a different threshold level VRl - V4. Each comparator increments a respective counter 22c-22f.

Parameter G: Peak amplitude A peak detector circuit 21g, e.g. a 1000 pF capacitor which is charged quickly through a precision rectifier, stores the highest positive peak of the pen velocity waveform. The high impedance discharge path (the reverse impedance of the rectifier) ensures that the capacitor voltage remains reasonably constant after the end of the signature until a reset operation is performed. The voltage on the capacitor is applied to an analogue-to-digital converter 22g.

Parameter H: Integration An integrator 21 L, e.g. a 1000 juF capacitor charged slowly at a rate dependent on the positive input voltage from the coils, is connected to the sensing unit. The capacitor voltage is amplified and converted to digital in an A/D converter 22h. This provides a reading which is proportional to the integral of the pen velocity waveform and thus represents the total positive component of the distance travelled by the pen.

The measurements described above have been confined only to the positive going component of the pen velocity waveform. It is apparent that similar measurements may be made to the negative going component either in addition to or in place of those described above. The outputs from the measuring circuits are all in digital form, e.g. a set of binary encoded numbers, which may then be compared with the upper and lower limit reference values extracted from the store. The design of logic circuits to accomplish such digital comparisons is readily achieved by those skilled in the art and needs no detailed explanation here. As an example of the requirements for the logic circuits let us assume that the result of each measurement is presented in the form of a binary coded decimal number between zero and 99.Thus the results of the eight measurements described above would be transmitted a group representing the signature code. If it is assumed that the signature Duration is 3.7 seconds and the pen makes 6 countable contacts in that time the output numbers for measurement A 8 B can be 37 and 06 respectively. Let us assume that the numbers for the remaining six measurements are respectively 33, 44, 55, 66, 77 and 88.

The combined code would therefore become 3706334455667788. The signature would be validated by comparing this sequence with the appropriate reference limits extracted from the store.

The reference limits held in the store must have been established previously by obtaining from each signatory a number of sample signatures. It will be found that the average signatory will not reproduce exactly the same reference code over a number of signatures, there will be variations. However once a person has produced a few signatures, say six, it will be fairly easy to decide what the upper and lower reference limits for each parameter should be. For example, it may be found that a particular person's signature may have a duration varying from 3.1 to 3.9 seconds.

The limits could be set to 31 and 39 respectively and subsequent signatures must have a duration not exceeding these limits to be accepted as valid.

Although the signature code has the appearence of a 16-digit decimal number it cannot of course be treated as such in binary terms, bearing in mind that the two digit number for each parameter must be compared separately. In practice the code would be produced as a binary sequence of eight 8-bit bytes, each byte being the result of one measurement and being the binary coded decimal two-digit number representing that measurement.

Upon arrival at a central processor, each byte would be compared with two bytes of stored information defining the limits for that parameter, the successive bytes being identified by their order of arrival. At the end of the sequence, the stored results would be processed to reach an accept/reject decision for transmission back to the point at which validation is required. At its simplest, any one byte falling outside the defined limits could trigger the reject signal.

Using the above figures, the central processor must provide a minimum of 1 6 bytes of storage for each signature. At the expense of somewhat greater complexity, a 'guard space' might be provided so that, in the event that a given signatory, over a period of time, persistently produced signatures falling close to one of the test limits, that limit would automatically recede so as to maintain a predetermined mean guard space. Conversely, if a limit were rarely approached, that limit might advance so as to restore the mean guard space. This could accommodate the gradual changes that occur in a person's signature with the passage of time.

An interesting consequence of the mechanism described in that, since narrow passbands confer greater immunity from unauthorized access, the signatory has the opportunity to improve his own security by cultivating an accurately repeatable signature.

The effectiveness of the system may be judged according to two criteria: (i) The ability to reject an invalid waveform.

(ii) The ability to accept a valid waveform.

These criteria may be redefined as: (i) Ff, the failure rate for rejection of invalid waveforms, i.e. the probability that an invalid waveform will be validated.

(ii) Fv, the failure rate for acceptance of valid waveforms, i.e. the probability that a valid waveform will be rejected.

Ideally both F, and Fv should be zero, but this is only possible if all samples of a given waveform are identical. In the case of signature or speech waveforms this obviously will not be the case. Therefore in a practical system arbitrary values may have to be set for Ff and Fv, e.g. 10% and 1% respectively. The difference in these values in respect of a signature validation system for, say bank account holders may be justified on the basis that it is more important for the bank to avoid annoying genuine account holders than to provide absolute security against forgery.

If, as previously described, the first half dozen or so samples of a person's signature are taken as specimens to determine his validation passbands, against which his subsequent signatures are compared, the system is able to detect forgeriesFf is very low-but it is not tolerant of the normal variation in a given person's signature. There is a high probability that a genuine signature will be rejected, i.e. Fv is too high.

A direct trade-off between Ff and Fv may be arranged by extending the limits derived from the specimen samples by, say, 10%. The wider passbands so produced will allow more signatures, both valid and invalid, to be accepted, hence Fv decreases and Ff increases.

This arbitrary extension of the limits is a useful resort when only a few master samples of a given person's signature are available, since the person might not, by then, have revealed his full range of normal variation.

However, it would be practical to add every subsequent example of that signature, if accepted as valid, to the original set of master samples, making a progressive adjustment of the limits in the light of the accumulating data.

As an alternative to the signature sensing arrangement shown in Fig. 3 the following arrangement may be considered. A single layer of closely-spaced conductors is made to cover the whole area of the signature tablet immediately beneath the pressure-sensitive paper and hence in close proximity to the magnetic pen tip when this is in paper contact. The centre-to-centre spacing of adjacent conductors is made approximately equal to the diameter of the pen tip. To detect vertical excursions of the pen the conductors would be made to run horizontally across the tablet.

For a given direction of pen movement, all even-numbered conductors would be connected in series to give a positive-going output, while all odd-numbered conductors would be similarly connected to give a negativegoing output. This arrangement has several opertional features of interest. If the two sets of conductors are further connected in series, they will behave, for distant fields, as a selfcancelling bifilar winding. They will thus produce no resultant output in response to interferring external fields. Similarly, it is to be expected that the magnetic pen field will also be ignored while the pen remains poised above the paper, since the flux diverging from its tip will inevitably embrace more than one of the opposing conductors. However, when the pen is lowered so as to contact the paper, the flux concentrated at its tip will mainly interact with only one conductor at a time.

Any fringe flux, and flux returning over a less concentrated path to the pen magnet, will again link more than one opposing conductor and produce no resultant output.

Consider now, the effect of moving the pen, in contact, over successive conductors in turn.

If a positive-going output results from the passage of the pen over a given conductor, its continued motion over the adjacent conductor will result in a negative-going output, and so on. The resultant output from the composite winding, in response to a continuous pen movement, will be alternating voltage having a frequency and amplitude proportional to pen velocity. If the pen is removed from contact with the paper, to a distance exceeding the spacing between adjacent conductors, this wa- veform will cease, making it a suitable substitute for the 'pen contact' signal generated by mechanical switch contacts.

Simple rectification of this alternating voltage will produce a unidirectional waveform having an instantaneous amplitude proportional to pen velocity capable of analysis by existing circuits. In this simple form, the output would no longer indicate the direction of pen movement, since it would retain the polarity determined by the rectifier regardless of whether the pen was moving up or down the tablet. This need be no disadvantage, since the resultant waveform would still be characteristic of the specific signature. In fact, it might contain more inforr ration since the circuits previously described examine only the positive-going half-cycle of the present penvelocity waveform.

This tablet system is open to a number of variants. For instance, the frequency of the waveform, rather than the amplitude, might be made the measure of pen velocity. Directional information might be re-introduced by adopting a four-phase configuration for the conductors. Alternatively, the two sets of conductors might be arranged, by the subsequent circuitry, to appear in series opposition for the generation of an a.c. 'pen contact' waveform, but in series aiding for the generation of a pen-velocity waveform similar to that at present in use. In this latter arrangement, it would be necessary to provide a low-reluctance path to steer the return flux clear of the remoter conductors on the tablet, otherwise overall cancellation would occur. To this end, it might be advantageous to place the magnet system under the tablet and use only a highpermeability soft magnetic material as the pen tip.

There remains the more elaborate possibility of exploiting the alternating pen-velocity waveform by observing the number and timing of the individual pulses generated. This would describe the pen movement in much greater detail.

Claims (16)

1. Apparatus for validating a complex analogue waveform comprising a plurality of individual parameter measuring means each of which is arranged to measure a predetermined different parameter of the waveform, a same number of individual storage means each of which is arranged to hold upper and lower limit reference values for a different one of the parameters being measured, and a same number of comparator means each of which is arranged to compare the output of one measuring means with the stored reference values for the parameter being measured by that measuring means to provide an output indicative of whether or not the measured parameter value is within said limit values.

2. Apparatus according to claim 1 including a library of storage locations, one for each of a number of different complex waveforms, each storage location comprising a number of portions one of which holds an identification address within the library for the waveform, the remaining portions each storing the upper and lower limit reference values for one parameter of the waveform, and means for extracting from an addressed library location the stored reference values for transfer to the individual storage means for comparison with the output of the measuring means.

3. Apparatus according to claim 2 including means for generating the complex waveform and means for generating a library address for the waveform.

4. Apparatus according to claim 3 in which the waveform generating means com-: prises means for sensing movements of ? stylus on a writing surface to produce electrical signals related to such movements.

5. Apparatus according to any preceding claim in which the measuring means include means for measuring the duration of the waveform.

6. Apparatus according to any preceding claim in which the measuring means include means for digitally counting the number of gaps, if any, in the waveform.

7. Apparatus according to any preceding claim in which the measuring means include means for measuring the number of zero crossings in the waveform.

8. Apparatus according to any preceding claim in which the measuring means include means for measuring the number of times the waveform exceeds a predetermined threshold level.

9. Apparatus according to claim 4 or any one of claims 5-8 when dependent on claim 4 wherein the sensing means comprises a writing surface, a magnetic stylus for writing on the writing surface, and electromagnetic induction means disposed beneath the writing surface to produce an induced e.m.f. in response to the writing movements of the stylus.

10. Apparatus according to claim 9 in which the electromagnetic induction means comprises a pair of coils mounted vertically with respect to the writing surface and parallel to the normal direction of writing, the two coils being at the upper and lower edges respectively of the writing area, the two coils being electrically connected in series.

11. Apparatus according to claim 9 or 10 wherein the sensing means includes electrical switch means responsive to contact pressure of the stylus on the writing surface.

1 2. Apparatus according to claim 11 in which the duration measuring means comprises an astable multivibrator the output of which is fed via a switch to a digital counter, the switch being responsive to an electrical signal from the sensing unit when the stylus is in contact with the writing surface.

13. Apparatus according to claim 9 or 10 wherein the threshold exceeding measuring means comprises a comparator to which the output of the electromagnetic induction means is applied via an amplifier together with a predetermined reference signal, the output of the comparator being applied to a digital counter.

14. Apparatus according to claim 9 or 10 wherein the measuring means include means for measuring the highest peak of a given polarity of the waveform and means for converting this measurement into a digital representation.

15. Apparatus according to claim 9 or 10 wherein the measuring means include means for deriving a signal proportional to the inte gral of the stylus velocity waveform for the duration of the waveform and means for con verting this measurement into a digital representation.

16. Apparatus according to claim 9, 10 or 11 wherein the measuring means include means for deriving a signal proportional to the integral of the stylus velocity waveform for the duration of the waveform and means for converting this measurement into a digital representation.

1 7. Apparatus for validating a complex analogue waveform substantially as described with reference to the accompanying drawings.

16. Apparatus for validating a complex analogue waveform substantially as described with reference to the accompanying drawings.

CLAIMS (2 Feb 1980)

11. Apparatus according to claim 9 in which the electromagnetic induction means comprises a single layer of parallel conductors covering the whole area of the writing surface, the conductors being dimensioned with centre-to-centre spacing approximately equal to the diameter of the magnetic stylus and electrically connected in series.

12. Apparatus according to claim 9, 10 or 11 wherein the sensing means includes electrical switch means responsive to contact pressure of the stylus on the writing surface.

1 3. Apparatus according to claim 1 2 in which the duration measuring means comprises an astable multivibrator the output of which is fed via a switch to a digital counter, the switch being responsive to an electrical signal from the sensing unit when the stylus is in contact with the writing surface.

14. Apparatus according to claim 9, 10 or 11 wherein the threshold exceeding measuring means comprises a comparator to which the output of the electromagnetic induction means is applied via an amplifier together with a predetermined reference signal, the output of the comparator being applied to a digital counter.

1 5. Apparatus according to claim 9, 10 or 11 wherein the measuring means include means for measuring the highest peak of a given polarity of the waveform and means for converting this measurement into a digital representation.