JPH0370239B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to an individual verification device capable of verifying and identifying a pre-registered individual according to an input voice pattern, for example.

[Technical background of the invention and its problems] In financial operations and building access control operations, verification and verification of individuals is extremely important for crime prevention. Currently, the ID card method used in banking, for example, is widely known as a typical personal verification method. This ID card method verifies an individual by comparing a pre-registered ID number magnetically recorded on the ID card with an ID number newly entered using a numeric keypad, and its logical structure is simple. , is widely used because it can perform reliable verification.

However, this ID card method has the disadvantage that the above verification is practically impossible if the ID card is lost or the registered ID number is forgotten. In addition, if the ID number recorded on the lost ID card becomes known to a third party by some means, this could lead to incorrect personal identification and unauthorized use of the ID card. There are also.

Therefore, recently, various attempts have been made to verify individuals using characteristic patterns such as voice, hand prints, handwritten characters, etc. unique to individuals, instead of the above-mentioned ID numbers. However, since the personal matching process is only performed by a simple method such as exact matching with a registered pattern or feature value determination, a sufficiently high matching rate has not been achieved. Therefore, there are still some problems in practical application.

[Purpose of the invention]

The present invention has been made in consideration of these circumstances, and its purpose is to easily and reliably identify individuals who have been registered in advance according to an input pattern and perform personal verification. The purpose of the present invention is to provide a highly practical personal verification device that can perform the following tasks.

[Summary of the invention]

In the personal verification device according to the present invention, a first residual is obtained by removing components of a standard pattern for a key code from a pattern of voice, etc. input by each individual (registered input pattern) for a key code determined for each individual. A pattern is generated and registered in advance as a personal matching pattern.

When an input pattern such as voice is input by the person to be verified, the category of the input pattern is recognized,
A second residual pattern is generated by removing the standard pattern of the same category from the input pattern.
Then, this second residual pattern is compared with the previously registered pattern for personal matching to perform personal matching of the person to be matched.

〔Effect of the invention〕

Therefore, according to the present invention, since the pattern used for the matching process has strong individuality from which standard pattern components have been removed, it is possible to perform extremely effective and reliable personal matching. Become. Moreover, since a highly individualized pattern is obtained through the simple process of removing standard pattern components from the registered input pattern, the verification process can be made more reliable, which has great practical advantages.

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of an embodiment device, which includes a voice recognition device 1, a personal verification device 2, a voice response device 3,
It is constituted by a system control device 4 that controls these processing operations. A voice signal input from the microphone 5 is input to the voice recognition device 1 and subjected to recognition processing. This speech recognition device 1 converts the input speech signal into, for example, a band pass signal.
The voice input unit 11, which is composed of a group of filters, performs frequency analysis or the like and converts it into characteristic parameters. The feature parameters of the audio signal converted by the audio input unit 11 are subjected to processing such as audio interval detection and time axis normalization via the preprocessing unit 12, and then transferred to the recognition unit 13. . The recognition unit 13 recognizes the category of the voice by calculating the similarity between the standard parameters (feature patterns) of the voice registered in advance in the dictionary storage unit 14 and the characteristic parameters of the input voice. . Then, the voice code string signal which is the result of this recognition is given to the personal verification device 2 together with the voice characteristic pattern.

In the personal verification device 2, the voice feature pattern is input to the personal feature extracting section 16, and the standard voice pattern indicated by the code string signal is read from the dictionary storage section 14 and used as the voice feature pattern. It is more subtracted. As a result, only individual characteristic components of the input voice can be obtained as a residual pattern (second residual pattern). This residual pattern is supplied to the personal matching section 17, and is compared with the personal matching pattern registered in advance in the personal matching dictionary storage section 18. The individual matching pattern registered in the dictionary storage unit 18 is a residual pattern (first (residual pattern). This personal verification pattern is also obtained through the processing by the voice confirmation device 1 and the processing by the personal feature extraction section 16 described above. This personal verification result is transmitted from the voice response device 3 to the speaker 6.
A voice response is provided via.

Now, in this apparatus configured as described above, dictionary registration of a personal matching pattern and personal matching processing according to an input pattern using the personal matching pattern registered in the dictionary are performed as follows.

Now, suppose that the pattern of sounds, characters, etc. of the key code used for personal verification is represented by a vector f ^(l) . Here, l indicates the category (l=1.2.3 to L) to which the above pattern belongs. On the other hand, the standard pattern ^(l) of category l is, for example, a set of patterns of category l obtained from an unspecified number of people {f _j ^(l)
(x _i )}. However, the above j is (j=
1.2.3…Jl). This standard pattern
^(l) is created from the set of patterns {f _j ^(l) (x _i )} in the same way as standard patterns are created in the conventional composite similarity method or _mixed similarity method ^. x _i )} characteristic kernel K _(xi,xi ′ ₎ = 〓 ^j f _j(xi) Let the eigenvector of f j _(xi) be {φ ^(l) _n }, ^(l) = {φ ^(l) _n } (m=1.2.~M _l ), which indicates the average pattern or representative deformed pattern of a given set of patterns {f ^(l) _j }, and is the least squares It is expressed as an optimal approximation in the sense of and,
The standard pattern {φ ^(l) _n } is the same category l
It is given as a dictionary that has an orthogonal relationship with each other.

Now, the voice pattern of the key code that a specific individual α registers for personal verification is shown as gα ^(l) . This voice pattern gα ^(l) is input by the utterance of the individual α, and is a category-l pattern that has a unique distortion indicating the characteristics of the individual α. Therefore, in this device, the average pattern component φ ^(l) _n , which is a standard pattern of the same category l mentioned above, is removed from the individual's speech pattern gα ^(l) , and a residual is created that emphasizes the individual characteristics mentioned above. pattern
hα ^(l) is obtained and registered in the dictionary in the storage unit 18 as a personal matching pattern. That is, the residual pattern hα ^(l) is obtained, for example, as follows.

However, in the above formula, (.) indicates inner product processing of pattern vectors, and ‖ ‖ indicates norm. The denominator term shown in this equation is introduced for normalization, and can be set to 1 to simplify the calculation. In this way, individual matching patterns are obtained based on the registered input patterns inputted by each of the key codes respectively defined by a plurality of individuals, and are stored in the personal matching dictionary storage section 18, classified by category, for example. Registered in the dictionary.

Now, as mentioned above, each individual's matching pattern h ^(l)
is registered in the dictionary in advance, and then personal verification of the person to be verified is performed as follows. When the person to be verified inputs a pattern P by voice or the like, the component removal of the standard pattern described above is performed on this pattern P in the same manner as described above, and the residual pattern q ^(l) is obtained.

However, it goes without saying that this process is performed using the standard pattern φ ^(l)/m of the same category l as the input pattern P described above. However, personal verification section 17
The residual pattern q ^(l) obtained for this input pattern P is successively compared with the matching pattern h ^(l) for each individual registered in advance in the dictionary storage unit 18 to perform individual matching. I do. This verification is performed using a known simple similarity method, complex similarity method, or the like as appropriate.
For example, the similarity S(q ^{( l )} ,
hα ^(l) ) can be obtained as follows.

S (q ^(l) , hα ^(l) ) = (q ^(l) , hα ^(l) /‖q ^(l) ‖・‖
hα ^(l) ‖ Then, the personal matching judgment according to the similarity result obtained in this way is performed under the condition that S(q ^(l) , hα ^(l) ) ≧ θ ₁ , where the judgment threshold is θ ₁ . This is performed assuming that the person to be verified is the individual α, and the person to be verified is not the individual α under the condition that S(q ^(l) ,hα ^(l) )< _θ1 .

Furthermore, if there are a plurality of matching patterns hα ^(l) registered in the dictionary, that is, if there are a plurality of individuals who have defined the same category, a similarity value is obtained for each of them. Then, the maximum similarity S ^(l) _nax that indicates the maximum (or minimum) value among them,
The next similarity S ^(l) _oext is calculated, and S ^(l) _nax = Max {S (q ^(l) , hα ^(l) ) | α} > θ ₂ , and S ^(l) _nax -S ^(l) _oext > θ ₃ may be determined, and only when these conditions are satisfied, the person to be verified is determined to be the individual α. However, the above θ ₂ and θ ₃ are determination thresholds each consisting of a predetermined constant.

As described above, according to the present device, since individual matching is performed using the residual pattern between a highly individualized input pattern and a standard pattern, the matching accuracy is extremely high. Moreover, the components of standard patterns that share the same category are removed from the input pattern, and individual matching is performed by searching for residual patterns with strong individuality.This method effectively utilizes individual characteristics to perform matching. There is no possibility of obtaining erroneous verification results, and its practical advantages are enormous. Furthermore, since the above-mentioned processing is not so complicated, there is also an effect that the apparatus can be easily realized.

Incidentally, the system control device 4 that executes and controls the above verification process may execute the control process as shown in FIG. 2, for example.

In other words, if the key code set by the individual is, for example, 0
It is assumed that it is given by n codes selected from ~9, A, A~. In this case, first, as shown in step 21, an n-digit code is voiced and input, and in step 22 the voice is recognized and converted into an n-digit voice code. The speech recognition device 1 is caused to perform this process. However, the control device 4
In step 23, the above n
It is determined whether the keyword indicated by the digit voice code is registered for personal verification. If the voice code is not registered, it is rejected, and if it is registered, the process proceeds to the next step. Then, in step 24, the individual characteristic parameters (residual pattern) are stored in the individual characteristic extraction unit 16.
After that, step 25, 2
At step 6, dictionary search and collation processing are completed. This matching process is performed by similarity calculation, etc., as described above. Then, in step 27, the above-mentioned conditional determination of the verification result is performed to determine the identity of the person to be verified. In the rare case that an individual cannot be identified by this, this will be rejected.
As a result of this reject, the above-described verification process is retried, input procedure guidance, etc. are performed.

As explained above, the control flow of the verification process in this apparatus is also very simple, and no special verification procedure is particularly required. Furthermore, known methods such as the similarity method can be used as they are. Furthermore, simply by increasing the types of patterns used for matching, it is possible to perform even more accurate matching.

Note that the present invention is not limited to the above embodiments. In the embodiment, the matching pattern is provided as a voice, but it may also be a handwritten character or the like. in this case,
A tablet or the like may be used as the input device, and feature parameters may be provided as features related to character line segments and their densities. Furthermore, the method of the matching method itself, that is, the similarity calculation method, etc. may be determined according to the specifications. In short, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of drawings]

FIG. 1 is a schematic diagram of an apparatus according to an embodiment of the present invention;
FIG. 2 is a diagram showing an example of a process control flow in the embodiment apparatus. 1...Voice recognition device, 2...Personal verification device, 3
...Voice response unit, 4...System control device, 5...
. . . Microphone, 6 .

Claims (1)

[Scope of Claims] 1. Means for generating a first residual pattern by removing components of a standard pattern for a key code from a pattern input by each individual regarding a key code determined for each individual; A registration means for registering the generated first residual pattern as a personal matching pattern, a means for confirming the category of the input pattern input by the person to be matched, and a standard of the same category as the category recognized by this means. means for generating a second residual pattern by removing the pattern from the input pattern; and a means for comparing the second residual pattern generated by the means with the personal matching pattern registered by the registration means. 1. A personal verification device comprising means for personally verifying a verification person.