JP3733668B2 - Method and apparatus for creating signature verification template and signature verification apparatus - Google Patents

Description

【００５６】
なお、上式において、ｍ１，ｍ２は、夫々、テンプレート評価に用いる正例サインデータ及び負例サインデータの数を表す。そして、上式では、正例サイン及び負例サイン夫々に対する照合誤差の平均が、適応度Ｆとして算出され、この適応度Ｆが１に近いほど、生成したテンプレート（換言すればテンプレートの元となった遺伝子座）は正例サインの個人性特徴を反映していることになるので、Ｓ４２０では、適応度Ｆが最も１に近い染色体内の遺伝子座をエリート遺伝子座として選別する。
【００５７】
また、図７に示す照合ネットワークは、前述のサイン照合部２２の機能を表しており、染色体を構成する複数の遺伝子座から生成したＸサインデータ及びＹサインデータに対する複数のメンバシップ関数からなるテンプレートと、このテンプレートを構成する各メンバシップ関数と照合対象となるサインデータにおける各メンバシップ関数に対応した部分曲線毎のＸサインデータ及びＹサインデータとから得られる各部分曲線毎の一致度をかけ算して、照合対象となるサインデータとテンプレートの作成に用いた正例サインデータとの照合率を求めるかけ算部とから構成される。つまり、テンプレートの評価にはサイン照合時と同様の照合ネットワークが使用され、評価すべきテンプレートを用いて、正例及び負例サインデータとテンプレートの作成に用いた正例サインデータとを実際に照合させるのである。そして、上記生成したメンバシップ関数を用いれば、部分曲線同士の一致度が高い程、その値が１に近づき、得られる照合率も１に近づくことになる。
また上記Ｓ３８０〜Ｓ４２０の一連の処理は、Ｓ３７０にて選定した染色体の中の一つの遺伝子を増殖／突然変異させて得られるｍ個の遺伝子座を夫々有するｍ個の染色体に対するｍ個のテンプレートを生成して、各テンプレートの評価を行うことにより、ｍ個の遺伝子座の中からエリート遺伝子座を選別するための処理であるが、これら一連の処理は、ｍ個の遺伝子座を夫々有するｍ個の染色体に対するｍ個のテンプレートを同時に生成して、各テンプレートの評価を同時に行うようにしてもよく、或いは、Ｓ３７０にて選定した染色体からテンプレートを生成してこれを評価し、次にＳ３８０にて選定した任意の遺伝子座を突然変異させて、その遺伝子座を有する染色体からテンプレートを生成してこれを評価し、更に先に突然変異させた遺伝子座を再度突然変異させて、その遺伝子座を有する染色体からテンプレートを生成してこれを評価する、といった手順で、選定した一つの遺伝子座を順に突然変異させて、その遺伝子座に対応したテンプレートを順に評価することにより、エリート遺伝子座を選別するようにしてもよい。
【００５８】
そして、上記のようにエリート遺伝子座が選別されると、Ｓ３７０にて選定した染色体を構成する全ての遺伝子座に対する世代更新（つまりエリート染色体の選定）は完了したか否かを判断し（Ｓ４３０）、全ての遺伝子座に対する世代更新が完了していなければ、Ｓ３８０に移行する。するとＳ３８０では、現在遺伝子操作の対象となっている染色体を構成するｉ個の遺伝子座の内、上記遺伝子操作によってエリート遺伝子座が選別されていない任意の遺伝子座が選定され、続くＳ３９０以降の処理で、この選定された遺伝子座に対する遺伝子操作がなされて、エリート遺伝子座が選別される。この結果、Ｓ３７０にて選定した染色体を構成する全ての遺伝子座に対する遺伝子操作が行われ、その全ての遺伝子座に対してエリート遺伝子座が設定されることになる。そして、Ｓ４３０にて、遺伝子座の世代更新が完了したと判断されると、Ｓ４４０に移行し、Ｓ３７０にて選定された染色体を、上記選別したエリート遺伝子座からなる染色体に更新する。
【００５９】
また次にＳ４４０にて染色体を更新すると、上記Ｓ３７０〜Ｓ４４０の処理により、ｋ個の染色体全てを更新したか否かを判断する（Ｓ４５０）。そして、全ての染色体に対する更新が完了していなければ、再度Ｓ３７０に移行する。するとＳ３７０では、ｋ個の染色体の内、更新が完了していない任意の染色体が選定され、続くＳ３８０以降の処理で、この選定された染色体を構成する各遺伝子座に対する遺伝子操作及びエリート遺伝子座の選定がなされ、この染色体が更新されることになる。この結果、Ｓ３１０〜Ｓ３６０の処理により生成されたｋ個の染色体全てに対して世代交代がなされることになる。
【００６０】
次に、Ｓ４５０にて、ｋ個の染色体全てを更新したと判断されると、今度は、Ｓ４６０に移行して、上記のように更新したｋ個の染色体の中から、テンプレートの評価の低い順に所定割合（例えば、２０％）だけ削除する、所謂染色体淘汰のための処理を実行する。なお、削除する染色体の選定には、例えば、Ｓ３７０〜Ｓ４４０の処理によってｋ個の染色体を各々世代交代させた際に得られた評価結果（つまり適応度Ｆ）を使用することができる。
【００６１】
そして、このようにｋ個の染色体の中から、評価の低い染色体を所定割合だけ削除すると、Ｓ４７０にて、残った染色体の中から任意の２つの染色体を選択して、一方の染色体を構成する所定ストローク番号以上の遺伝子座と、他方の染色体を構成する所定ストローク番号未満の遺伝子座とを組み合わせることにより、新たな染色体を生成する、所謂染色体の交叉処理を行うことにより、Ｓ４６０で削除した染色体の数だけ新たな染色体を補充する。
【００６２】
またこのように染色体を補充すると、今度は、Ｓ４８０に移行して、その補充した染色体毎にテンプレートを生成し、続くＳ４９０にて、その生成したテンプレートの評価を行う。なお、テンプレートの生成及びその評価は、Ｓ４００及びＳ４１０と同様の手順で実行される。
【００６３】
そして、続くＳ５００では、Ｓ４９０の処理により得られた新たな染色体の評価結果と、Ｓ３７０〜Ｓ４４０の処理により得られた残りの染色体の評価結果とに基づき、ｋ個の染色体の中から、最も評価の高い（つまり適応度Ｆが最も１に近い）染色体をエリート染色体として選別する。そして、続くＳ５１０では、この選別したエリート染色体の適応度Ｆは、予め設定された判定基準（例えば、９５％）以上であるか否かを判断し、エリート染色体の適応度Ｆが判定基準以上であれば、エリート染色体はサイン登録者の個人性特徴を最適に反映しており、これに基づき生成したテンプレートを用いればサイン照合を問題なく実行できるものと判断して、当該遺伝子演算処理を終了する。
【００６４】
一方、Ｓ５１０にてエリート染色体の適応度Ｆは判定基準に達していないと判断された場合には、再度遺伝子操作を行って適応度が判定基準以上となるエリート染色体を生成すべく、Ｓ３７０に移行する。この結果、Ｓ３７０以降の処理では、遺伝子操作により更新された染色体と、淘汰，交叉により新たに生成された染色体とからなる、合計ｋ個の第２世代の染色体に対する遺伝子操作が再度実行されることになり、最終的には、適応度Ｆが判定基準以上となる最適なエリート染色体が選別されて、当該遺伝子演算処理が終了することになる。
【００６５】
なお、図３に示した遺伝子演算は、前述の論文に記載されたものと略同様であり、この論文に開示された技術と本実施例とは、染色体を構成する遺伝子座の構成，及びこの遺伝子座から生成されるテンプレートの構成が異なる。そして、遺伝子演算の手順は、図３に示したフローチャートの手順に限定されるものではなく、適宜変更することができる。
【００６６】
次に、図２に戻り、Ｓ１５０の遺伝子演算処理によって適応度Ｆが判定基準以上となるエリート染色体が選別されると、Ｓ１６０にて、このエリート染色体からサイン照合用テンプレートを生成する。
なお、Ｓ１６０では、エリート染色体からサイン照合用のテンプレートを生成するようにしても良いが、エリート染色体に対応したテンプレートは、Ｓ１５０の遺伝子演算処理実行時に既に生成されているので、そのテンプレートを読み込むようにしてもよい。また、Ｓ１６０にて生成されるテンプレートには、各部分曲線毎のメンバシップ関数に加えて、そのメンバシップ関数に対応した部分曲線を特定するために、複数の遺伝子座からなる染色体の情報も含まれる。
【００６７】
そして、このようにエリート染色体に対応したサイン照合用テンプレートが生成されると、今度は、Ｓ１７０に移行して、表示装置６に、サイン登録者にＩＤ番号を入力させるためのメッセージを表示し、サイン登録者がキーボード４を操作してＩＤ番号を入力するのを受け付ける、ＩＤコード入力処理を実行する。
【００６８】
そして続くＳ１８０では、Ｓ１６０にて生成又は読み込んだサイン照合用のテンプレートに、Ｓ１７０のＩＤコード入力処理により取り込んだサイン登録者のＩＤ番号を付与して、テンプレート記憶媒体２０に格納し、処理を一旦終了する。
【００６９】
以上のサイン登録時の演算処理は、一律ではないが、数十分から数分以下で完了する。そして、上記のように生成されたテンプレートのデータサイズは、一つのメンバシップ関数が１６バイトで構成されるものとすれば、通常は、約５００バイト程度の容量に収まる。
【００７０】
次に、サイン照合時にマイクロコンピュータ１０にて実行される演算処理を説明する。
既述したように、マイクロコンピュータ１０は、Ｓ１１０にてサイン登録か否かを判断し、サイン登録時でなければ（つまりサイン照合時には）、Ｓ２１０に移行する。そして、Ｓ２１０では、表示装置６に、照合対象者に自己のＩＤ番号を入力させるためのメッセージを表示し、照合対象者がキーボード４を操作してＩＤ番号を入力するのを受け付ける、ＩＤコード入力処理を実行する。
【００７１】
また、Ｓ２１０にて照合対象者のＩＤ番号を取り込むと、Ｓ２２０にて、このＩＤ番号に対応したテンプレートを、テンプレート記憶媒体２０から呼び出し、続くＳ２３０にて、表示装置６に、照合対象者にサインを記入させるためのメッセージを表示し、照合対象者がタブレット２にサインを筆記することによりタブレット２から入力されるサインデータを読み込む、サイン入力処理を実行する。
【００７２】
そして、このサイン入力処理により、照合対象者のサインデータが得られると、今度はＳ２４０に移行して、そのサインデータを正規化し、Ｓ２５０にて、その正規化後のサインデータとテンプレートとから、今回手書き入力されたサインとテンプレートの生成に用いた正例サインとを照合するためのマッチング演算を行う。なお、このマッチング演算は、先に図７を用いて説明したように、Ｓ２４０にて正規化したサインデータの中から、テンプレートを構成するメンバシップ関数に対応した部分曲線のデータを取り込み、そのデータとテンプレート作成時のデータとの一致度を、対応するメンバシップ関数を用いてファジー推論し、更にこのファジー推論により得られた各部分曲線毎のデータの一致度をかけ算することにより手書きサインと登録サインとの照合率を求める、といった手順で実行される。
【００７３】
そして、このようにＳ２５０のマッチング演算によって照合率が求められると、続くＳ２６０にて、照合率が、照合対象者をサイン登録者本人であると認証しうる判定値（例えば０．７）以上であるか否かを判断し、照合率が判定値以上であれば、照合対象者はサイン登録者本人であると認証し、逆に照合率が所定値に達していなければ、照合対象者にサインを再度入力させるか或いは照合対象者は偽者であるかを判断する、照合判断処理を実行する。そして、Ｓ２７０では、この照合判断処理による判断結果を、外部装置に出力し、当該処理を一旦終了する。
【００７４】
以上説明したように、本実施例のサイン照合装置によれば、遺伝的アルゴリズムに従い、登録者が記述した正例サインの中から、登録者の個人性特徴を最も反映した部分曲線を抽出し、この部分曲線毎に一致度をファジー推論するためのメンバシップ関数を生成することにより、サイン照合用のテンプレートを生成するようにしている。また、部分曲線を特定する遺伝子座は、部分曲線が属するストロークの番号と、そのストローク上での部分曲線の始点及び終点を表す３つの情報から構成されており、各メンバシップ関数は、この遺伝子座に対応して、部分曲線の長さに対応したＸ方向及びＹ方向のサインの位置変化量から構成されている。
【００７５】
このため、本実施例のサイン照合装置によれば、サイン照合時に、登録サインの個人性特徴を最も反映した部分曲線を利用して、登録サインと手書きサインとの一致度をファジー推論することができ、サイン照合の精度を確保することができると共に、サイン照合に用いるテンプレートの容量（換言すればメンバシップ関数の数）を少なくして、これを記憶するテンプレート記憶媒体２０の容量を小さくすることができる。
【００７６】
以上、本発明の一実施例について説明したが、本発明は上記実施例に限定されるものではなく、種々の態様をとることができる。
例えば、上記実施例では、サイン登録及びサイン照合を一つのサイン照合装置で実現するようにしたが、サイン登録時には、遺伝的アルゴリズムに従った演算処理を高速に行うために、専用のコンピュータを用い、サイン照合時には、サイン照合用の端末装置等にサイン登録された記憶媒体を組み込んでサイン照合を行うようにしても良い。
【００７７】
また、上記実施例では、染色体を構成する遺伝子座の数は、５〜５０個の範囲内で乱数等を用いて適宜設定されるものとして説明したが、この場合、最終的に得られるエリート染色体の遺伝子座の数も適宜設定されることになり、このエリート染色体から生成されるテンプレートのデータサイズもその都度変化することになる。このため、例えば、テンプレートのデータサイズを一定にしたい場合は、エリート染色体を選定する際の評価項目の一つに、遺伝子座の数を追加し、エリート染色体として、遺伝子座の数が所定値の染色体を選別するように構成するとか、或いは、染色体を構成する遺伝子座の数を所定値に固定するようにすればよい。そして、このようにすれば、テンプレートのデータサイズを常に一定にすることができる。
【００７８】
また上記実施例では、生成したテンプレートの評価のために、サイン登録者以外の者が記述した負例サインデータをサンプリングするものとして説明したが、例えば、乱数を発生させて、振幅を制限するフィルターを通過させたものに、正例サインデータを掛け合わせることにより、負例サインデータを生成する、といった手順で、正例サインデータから負例サインデータを生成するようにしてもよい。
【図面の簡単な説明】
【図１】 実施例のサイン照合装置全体の構成及び機能を表すブロック図である。
【図２】 実施例のサイン照合装置を構成するマイクロコンピュータにてサイン登録時及びサイン照合時に実行される演算処理を表すフローチャートである。
【図３】 図２のＳ１５０において実行される遺伝子演算処理の詳細を表すフローチャートである。
【図４】 タブレットから入力される時系列データの一例を表す説明図である。
【図５】 サイン登録時にタブレットから入力された時系列データに基づき生成される遺伝子座及び染色体を説明する説明図である。
【図６】 遺伝子座にて特定される部分曲線の一致度をファジー推論するためのメンバシップ関数の生成手順を説明する説明図である。
【図７】 サイン登録時に生成されるテンプレートの構成及びこれを用いてサイン照合を行う照合ネットワークの機能を説明する説明図である。
【符号の説明】
２…タブレット ４…キーボード ６…表示装置（ＬＣＤ）
１０…マイクロコンピュータ １２…信号処理部 １４…遺伝子操作部
１６…テンプレート生成部 ２０…テンプレート記憶媒体
２２…サイン照合部 ２４…判定出力部 ２８…制御部[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a template creation method and apparatus suitable for fuzzy inference of the degree of coincidence between a handwritten signature and a registered signature registered in advance, and a signature verification apparatus that performs signature verification using the created template. .
[0002]
[Prior art]
Conventionally, in a signature verification device that verifies a handwritten signature against a registered signature and authenticates that the person to whom verification is described is a registered person, a template for signature verification is created from the registered signature. The signature verification is performed by determining whether or not the handwritten signature matches or substantially matches the template.
[0003]
In other words, in signature verification, a correlation value (that is, a degree of single value) between a registered signature and a handwritten signature is obtained, and if this correlation value is high, the person who entered the signature by handwriting may be authenticated as the person himself. Using the image processing technique disclosed in Japanese Patent Application Laid-Open No. Hei 7-129770, feature points are extracted from an image of a registered sign, a template is created for the feature points, and handwritten input By comparing the signature feature point with the template, the degree of unification between the registered signature and the handwritten signature is obtained, or as described in JP-A-4-352080, The end part (Yes, Stop, Hari) is used as a feature point, and a membership function for fuzzy inference of the degree of coincidence between a handwritten signature and a registered signature is used as a template. By creating, performs signature verification by fuzzy inference, it has been made possible such.
[0004]
In such a conventional apparatus, the signature feature point is extracted and a template corresponding to the feature point is created. Therefore, compared to the case of creating a signature matching template from the entire signature, the template The amount of data can be reduced.
[0005]
[Problems to be solved by the invention]
However, conventionally, when extracting feature points from a registered signature, the personality feature of the signature is set in advance as a feature point such that the signature is the end of each successive stroke. For this reason, the template generated from the feature points does not necessarily reflect the personality characteristics of the sign registrant, and an error may occur in the accuracy of signature verification.
[0006]
In particular, in the apparatus disclosed in Japanese Patent Application Laid-Open No. 4-352080, a membership function for fuzzy inference of a single point of feature points is set as a template for signature verification. If the target feature point accurately reflects the personality characteristics of the sign registrant, the degree of singularity between the registered signature and the handwritten signature can be obtained with extremely high accuracy. Because the feature point that sets the value is fixed at the end of each stroke of the sine, even if the single value of the feature point is obtained by fuzzy inference, an error is likely to occur in the inference result of the single value There's a problem.
[0007]
The present invention has been made in view of such problems, and in creating a template for signature verification from the feature points of a registered signature, a portion that most reflects the personality characteristics of the signature registrant as the feature point. It is an object of the present invention to provide a signature collating template creation method and apparatus that can set a signature, and a signature collation apparatus that collates each signature using the created template.
[0008]
[Means for Solving the Problems]
According to the method for creating a signature verification template according to claim 1, which is made to achieve the above object, when creating a signature verification template, first, when the signature verification position, writing pressure, etc. The series data is taken in, the time series data is normalized by the size and position of the signature, and a plurality of partial curves representing the individuality characteristics of the signature are set from the normalized time series data. Then, for each partial curve, a membership function for fuzzy inference of the degree of coincidence with the signature data to be verified is generated, and the generated membership function for each partial curve is set as a template for sign verification. To do.
[0009]
Also, when setting the partial curve that is the target for generating the membership function that constitutes the template, first, the arbitrary time-series data is divided into a plurality of strokes divided into regions where the signs are continuous. By selecting, and selecting an arbitrary section on the selected stroke as a partial curve, a plurality of partial curves are initialized, and the plurality of partial curves are respectively assigned the stroke number and stroke to which the partial curve belongs. It is set as a gene locus composed of three types of information representing the start point of the partial curve above and the end point of the partial curve on the stroke.
[0010]
Next, using the locus group consisting of a plurality of loci representing each partial curve as the chromosome of the signature, the locus is manipulated according to a preset genetic algorithm to optimally express the personality characteristics of the signature. An elite chromosome is generated, and a partial curve specified at each gene locus constituting the generated elite chromosome is set as a partial curve for creating a template.
[0011]
That is, first, the method of the present invention is for creating a template for performing signature collation by fuzzy inference. A partial curve of a sine, which is a generation target of a membership function constituting the template, is represented by a genetic algorithm. Set according to.
[0012]
In addition, a method for setting a partial curve of a signature used for signature verification using a genetic algorithm as described above has already been proposed by the present inventors ("signature personality using a local improved genetic algorithm"). In order to set a partial curve representing the individual characteristics of a sign using a genetic algorithm, the “discovery of features”, IEEJ Transaction C, Vol. 116, No. 5 (1996) p548 to p555) is used. It is important to have a coding method that expresses “where is the personality characteristic of a sign”. As described in the above paper, all possible characteristics such as the length, inclination, angle, and curvature of the partial curve are displayed. When included in a gene locus, it takes time to set up a partial curve, and the amount of information on the set partial curve increases, so there are many membership functions for fuzzy inference. , It becomes large amount of data template.
[0013]
Therefore, in the present invention, when setting a partial curve of a sign using a genetic algorithm, the number of the stroke to which the partial curve belongs, the starting point of the partial curve on the stroke, and the partial curve on the stroke The end point (in other words, the length from the start point) is used, and each piece of information is used as a locus.
[0014]
As a result, it is possible to reduce the information amount of the partial curve necessary for setting the partial curve that most reflects the personality characteristics, and to shorten the time required for the setting. In addition, when creating a membership function from a set partial curve, it is only necessary to create a membership function based on the amount of data change corresponding to the length from the start point to the end point of the partial curve. The number of template data can be reduced.
[0015]
Here, as a genetic algorithm for setting a partial curve that most reflects the personality characteristics of the sign writer from the time series data of the signature as described above, a chromosome is constructed as described in claim 2. Individual loci of signatures are generated by multiplying and mutating arbitrary loci to generate multiple chromosomes with different loci and selecting the chromosome that best represents the individuality of the signature among the multiple chromosomes. It is only necessary to use a locally improved genetic algorithm that generates an elite chromosome by performing an operation for extracting a locus most suitable for representing each of the loci constituting the chromosome.
[0016]
In other words, in the conventional genetic algorithm known so far, elite chromosomes are generated by gradually changing the loci constituting the chromosomes in accordance with a predetermined procedure. However, in the locally improved genetic algorithm, as described above, an elite chromosome is generated by growing and mutating an arbitrary locus. Can be generated at a speed of about 10 times or more). Further, in the locally improved genetic algorithm, the locus is mutated, so that an elite chromosome closer to the optimal solution can be generated as compared with a general genetic algorithm in which the locus is gradually changed.
[0017]
In addition, when creating a membership function for fuzzy inference of the degree of coincidence between the registered signature and the handwritten signature for each partial curve set as described above, as described in claim 3, By taking time series data multiple times and obtaining the minimum and maximum values of the data change amount from the start point to the end point of each partial curve in the multiple time series data, the range from the minimum value to the maximum value is set. A trapezoidal membership function may be generated.
[0018]
In other words, the membership function that constitutes the template for sign verification is a trapezoidal membership function having a width from the minimum value to the maximum value of the data change amount from the start point to the end point of each partial curve in a plurality of time series data. If configured, the variation in each partial curve of time-series data (ie, sine) captured multiple times can be reflected in the membership function, and the degree of coincidence of each partial curve will affect the variation in sign writing during signature verification. It is possible to create a template that can be obtained with high accuracy without being performed.
[0019]
Next, in the signature verification template creation device according to claim 4, when the time series data at the time of writing the signature is input from the sign input means, the normalization means converts the time series data into the size of the signature. And the partial curve setting means sets a plurality of partial curves representing the personality characteristics of the signature from the normalized time series data, and the template generation means sets the partial curve for each of the set partial curves. Then, a membership function for fuzzy inference of the degree of coincidence with the signature data to be verified is generated, and the generated membership function for each partial curve is stored in a predetermined template storage medium as a template for signature verification. .
[0020]
When the partial curve setting means sets the partial curve, the partial curve initial setting means first selects an arbitrary stroke from a plurality of strokes obtained by dividing the normalized time-series data for each region where signs are continuous. By selecting and selecting an arbitrary section on the selected stroke as the partial curve, a plurality of partial curves are initialized, and the plurality of partial curves are respectively assigned to the stroke numbers to which the partial curves belong, It is set as a locus consisting of three types of information representing the start point of the partial curve on the stroke and the end point of the partial curve on the stroke.
[0021]
Then, the elite chromosome generating means operates the locus according to a preset genetic algorithm using the locus group consisting of a plurality of loci set by the partial curve initial setting means as a chromosome of a handwritten input signature. As a result, an elite chromosome that optimally represents the personality characteristics of the signature is generated, and the partial curve specified by each gene locus constituting the generated elite chromosome is set as a partial curve for template creation.
[0022]
That is, in the signature verification template creation apparatus of the present invention (Claim 4), the signature verification template is created according to the creation method described in Claim 1. Therefore, according to the apparatus of the present invention, in the elite chromosome generating means, the amount of information necessary for setting the partial curve most reflecting the personal characteristics of the signature is reduced according to the genetic algorithm, and it is necessary for the setting. The time can be shortened, and when the template generation means creates a membership function for each partial curve, the membership function is based on the data change amount corresponding to the length from the start point to the end point of the partial curve. Therefore, it is possible to reduce the number of membership functions and hence the amount of template data.
[0023]
Next, in the signature verification template creation device according to the fifth aspect, the elite chromosome generating means generates an elite chromosome according to the local improved genetic algorithm according to the second aspect. For this reason, the signature verification template creation device of the present invention (Claim 5) can create a signature verification template according to the invention method of Claim 2, and is similar to the invention method of Claim 2. The effect of can be obtained.
[0024]
Further, in the signature verification template creation device according to claim 6, the template generation means fetches the time-series data at the time of writing the signature a plurality of times via the sign input means, and the plurality of time-series data in the fetched time-series data By obtaining the minimum value and the maximum value of the data change amount from the start point to the end point of each partial curve, a trapezoidal membership function having a width from the minimum value to the maximum value is generated. For this reason, the signature verification template creation apparatus of the present invention (Claim 6) can create a signature verification template according to the invention method of Claim 3, and is similar to the invention method of Claim 3. Thus, it is possible to create a template that can obtain the degree of coincidence of each partial curve with high accuracy without being affected by variations in sign writing during signature verification.
[0025]
On the other hand, a signature verification apparatus according to a seventh aspect includes a template storage medium in which a template created by the creation apparatus according to any one of the fourth to sixth aspects is stored in advance, and is stored in the template storage medium. This is a device for collating a registered signature with a newly-written handwritten signature using the template that has been created.
[0026]
And in the signature collation apparatus of the present invention (Claim 7), the normalizing means normalizes the time series data at the time of sign writing input from the sign input means by the size and position of the signature, and the inference means From the normalized time series data, partial curve data corresponding to a plurality of membership functions stored as templates in the template storage medium are extracted, and registered from the membership functions corresponding to the data. The degree of coincidence of each partial curve between the signature and the handwritten signature is fuzzy-inferred, and the collation rate output means calculates the overall degree of coincidence between the registered signature and the handwritten signature from the degree of coincidence of each partial curve. The calculation result is output as a verification rate.
[0027]
Therefore, according to the signature collation apparatus of the present invention, the collation rate representing the degree of coincidence between the registered signature and the handwritten signature is obtained using the template created using the device according to any one of claims 4 to 6. For example, if this collation rate (value of 0 to 1.0) is 0.7 or more, it is determined that the person who entered the signature by handwriting is the registrant himself, and the collation rate is 0.3 or less. If it is present, it is determined that the person is a fake person, and if the collation rate is a value from 0.3 to 0.7, personal authentication can be performed by a procedure of inputting the signature again.
[0028]
According to the present invention, at the time of signature verification, time series data for each partial curve corresponding to the membership function constituting the template registered in the template storage medium is extracted, and the corresponding membership function is used. Thus, it is only necessary to fuzzy infer the degree of coincidence with the registered signature, and it is not necessary to analyze the shape of the entire signature at the time of signature verification, so that signature verification can be performed easily in a short time. Further, since the template is created for a partial curve that most reflects the personality characteristics of the signature, signature verification can be performed with extremely high accuracy.
[0029]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of the entire signature verification apparatus according to an embodiment to which the present invention is applied. In addition to the function as a signature verification device for verifying a registered signature and a handwritten signature using a previously registered signature verification template, the signature verification device of the present embodiment is a signature input by a signature registrant by handwriting. It also has a function as a signature verification template creation device that creates a signature verification template from the time-series data.
[0030]
As shown in FIG. 1, the sign collating apparatus of the present embodiment has an input surface for handwriting input of a signature, and sequentially detects the input position and writing pressure of the signature when writing the signature on the input surface. The tablet 2 as a sign input means for inputting the result, the keyboard 4 for inputting data and various commands to the signature verification device by external operation, the signature described in the tablet 2 and various messages from the device side A microcomputer composed mainly of a CPU, RAM, ROM, etc. that executes various arithmetic processes for sign registration and signature verification based on display devices 6 for display and input signals from the tablet 2 and keyboard 4 10.
[0031]
The display device 6 is for displaying the signature described on the tablet 2 in real time so that the sign writer can describe the signature while confirming on the display screen. For this, a thin liquid crystal display (LCD) mounted on the input surface of the tablet 2 is used.
[0032]
The microcomputer 10 executes arithmetic processing described later for sign registration and signature verification in accordance with a program stored in advance in a ROM or an external storage device (hard disk, optical disk, etc.). The arithmetic processing is represented by functional blocks. And as shown in FIG.
[0033]
That is, in the microcomputer 10, first, the time series data of the signature input from the tablet 2 is normalized by the signal processing unit 12 as normalization means. At the time of sign registration, the gene operation unit 14 serving as a partial curve setting unit sets a plurality of partial curves representing individual characteristics of the signature from the time-series data normalized by the signal processing unit 12, and further includes a template generation unit The template generation unit 16 generates a membership function for fuzzy inference of the degree of coincidence between the registered signature and the handwritten signature for each set partial curve, and uses this as a signature verification template. It is stored in the template storage medium 20 together with the registrant's ID number.
[0034]
At the time of signature verification, the signature verification unit 22 reads a template corresponding to the ID number of the person to be verified from the template storage medium 20, and compares the template with the time-series data of the signature normalized by the signal processing unit 12. The target person collates the handwritten input signature and the registered signature, and the determination output unit 24 determines whether the verification target person is the person who registered the signature from the verification result (collation rate). The result is output to an external device.
[0035]
The gene operation unit 14, template generation unit 16, template storage medium 20, signature verification unit 22, and determination output unit 24 are controlled by a control unit 28 that receives signals from the keyboard 4. The control unit 28 also performs display control for displaying a signature and various messages on the display device 6, and the control unit 28 also displays time-series data of the signature from the tablet 2 in order to display the signature on the display device 6. Entered.
[0036]
The template storage medium 20 can use a backup area of a RAM constituting the microcomputer 10, but for example, data storage such as a hard disk, a magneto-optical disk, an IC card, a memory card, a magnetic card, etc. An external storage medium such as a card having a function may be used.
[0037]
Next, arithmetic processing executed in the microcomputer 10 to realize the above functions will be described with reference to the flowcharts shown in FIGS.
The sign collating apparatus according to the present embodiment is configured to input a sign registration command by operating the keyboard 4 during sign registration. Therefore, the microcomputer 10 first determines in S110 (S represents a step) whether or not the signature is registered based on whether or not a signature registration command is input from the keyboard 4. The sign registration process is executed, and if it is not at the time of sign registration (that is, at the time of signature verification), the signature verification process after S210 is executed.
[0038]
In S120 executed at the time of sign registration, a sign written by the sign registrant on the tablet 2 (hereinafter referred to as a positive sign) is displayed a times (for example, 5 to 30 times) on the display device 6, and a person other than the sign registrant Displays a message to capture the signature written by imitating the signature registrant's signature (hereinafter referred to as negative signature) b times (for example, 10 to 85 times), and the positive signature or negative signature is described on the tablet 2 Thus, a signature input process for capturing the time series data of the signature input from the tablet 2 is executed.
[0039]
In S120, when the time-series data of the positive sign or the negative sign is captured, the time-series data is normalized by the size and position of the sign in S130, and one sign data for sign registration ( Positive sign data or negative sign data).
[0040]
In subsequent S140, a series of processes of S120 and S130 are executed n times (n = a + b), so that a total of n pieces of sine data including a positive example sine data and b negative example sine data. Is determined whether or not.
If n pieces of signature data have not been input, the processing of S120 and S130 is executed again. If n pieces of signature data have been inputted, the process proceeds to S150, and the inputted n signatures. Using the data, gene calculation is performed to set a plurality of partial curves that most reflect the personality characteristics of the sign registrant in the positive example signature.
[0041]
Note that the tablet 2 of this embodiment is configured to sequentially input coordinate data in the X and Y directions corresponding to the pen position at the time of writing a sign. 4 (b), two types of time-series data each representing a change in the position of the sign in the X-direction and y-direction (hereinafter, when these two types of time-series data are described separately, the X-direction Is obtained as X-sine data, and Y-direction time-series data is referred to as Y-sine data). However, the time series data shown in FIG. 4B represents X signature data obtained when the sign shown in FIG. 4A (illustrated as “tree” in the figure) is described in the tablet 2. . As is clear from FIG. 4, the time-series data in each direction obtained in S120 is generated for each stroke (1), (2),. There is no data between each stroke generated by raising.
[0042]
In this embodiment, the tablet 2 is described as being able to obtain time-series data representing changes in the X-direction and Y-direction of the sign. However, if a tablet capable of detecting writing pressure is used, the writing pressure can be determined. Time series data representing changes will also be obtained. When time series data of such writing pressure is obtained, a membership function for the writing pressure data may be generated at the time of generating a template to be described later.
[0043]
On the other hand, the normalization process executed in S130 is executed in order to absorb differences in the size of the tablet 2, the size of the signature at the time of writing, the inclination, etc. from the time series data of such a signature. Yes, specifically, normalization from the number of pixels in each direction of the tablet 2, or conversion of the described sign time-series data from coordinate data to vector data, etc., and sign data described under various different conditions Is corrected in the same way as described under the same conditions.
[0044]
Next, the gene calculation executed in S150 is a process for setting a plurality of partial curves in the positive signature according to a preset local improvement genetic algorithm, and is executed as shown in FIG.
That is, first, in S310, arbitrary positive example sign data is read from the positive example and negative example sign data (n pieces) captured and stored as described above. In the subsequent S320, an arbitrary section (that is, an arbitrary stroke) in which data is continuous in the read positive sign data is selected, and in S330, a gene representing an arbitrary section on the selected stroke is selected. Generate a locus.
[0045]
In subsequent S340, it is determined whether or not i loci have been generated by the processes of S320 and S330. If i loci have not been generated, the process proceeds to S320 again. As a result, i partial curves specified by i loci are set from one positive sign data.
[0046]
Note that an arbitrary value may be set for the number of loci i using random numbers or the like, but in this embodiment, the number of loci is 50 or more in consideration of later calculations, matching rates, and the like. The value of the number i is limited to be less than the number.
Further, S320 to S340 are processes corresponding to the partial curve initial setting means of the present invention (Claim 4). In S330, as illustrated in FIG. 5A, each sign selected in S320 is displayed. For each stroke (1), (2),..., Information indicating the stroke number ST and the start point p and end point i of any partial curve on the stroke (for example, 100% from the start point to the end point of the stroke) , A gene locus consisting of the starting point p and the end point e of the stroke) is set.
[0047]
However, in FIG. 5A, the sign is “tree”, and for each stroke {circle around (1)} to {circle around (4)} constituting this, a gene locus representing a partial curve on the stroke is generated. In practice, however, random numbers or the like are also used for selecting a stroke and setting a partial curve on the stroke, and a gene locus representing an arbitrary section of an arbitrary stroke is generated. Therefore, a plurality of partial curves may be set for the same stroke by a plurality of gene loci, and there are also strokes for which no partial curve is set.
[0048]
Next, when i loci are generated by the processes of S320 to S340, the process proceeds to S350, and the generated locus group consisting of the i loci is used as a chromosome of the positive signature. Store (see FIG. 5B). When the locus group consisting of i loci is stored as chromosomes in this way, in subsequent S360, it is determined whether or not k (for example, 50) chromosomes have been generated, and all the stored in S350 are determined. If the number of chromosomes has not reached k, the process proceeds to S320 again, and an arbitrary stroke is selected from the positive sign data read in S310, and the process of generating i loci is executed again. To do. As a result, in this embodiment, k chromosomes composed of i loci are generated from one positive example signature data.
[0049]
In this way, k chromosomes are generated, and when it is determined in S360, genetic manipulation is performed on each of the k chromosomes, and processing for changing generations of each chromosome at the locus level is executed. .
That is, first, an arbitrary chromosome to be genetically manipulated is selected from the k chromosomes generated and stored as described above (S370). When a chromosome is selected, an arbitrary gene locus is selected from i gene loci constituting the chromosome (S380), and the selected gene locus is expanded to m cells. The “m−1” loci thus made are mutated (S390). As a result, it is possible to obtain m chromosomes in which only one of the i loci differs from the mutated “m−1” loci and the original locus.
[0050]
In addition, the proliferation of a gene locus means copying information on a specific gene locus as it is. In addition, the mutation is to change the partial curve specified by the original locus into an arbitrary section of an arbitrary stroke, for example, to generate a random number and to mutate its value By multiplying the information, the partial curve corresponding to the gene locus may be randomly changed.
[0051]
When the growth / mutation of the gene locus is performed in this way, a template corresponding to a total of m gene loci of the original gene locus and the gene locus that has been propagated / mutated is created (S400). By evaluating each template (S410), an elite locus that most reflects the individual characteristics of the signature is selected from the m loci (S420).
[0052]
Here, for the generation of the template, among the n pieces of sign data sampled by the processing of S120 to S140, the positive sign data read in S310 and four of the remaining positive sign data are: A total of 5 positive sign data is used. When the template is actually generated, i pieces of chromosomes are formed from five pieces of positive sign data (specifically, X sign data and Y sign data constituting each positive sign data). The data corresponding to each partial curve specified at the gene locus is respectively read, and from the read data, the minimum value and the maximum value of the change amount in the X direction and the minimum value of the change amount in the Y direction in each partial curve A maximum value is obtained, and a membership function for X sine data and Y sine data is generated for each partial curve from the minimum value and maximum value of the change amount in each direction, and each membership function is used as a template. Set.
[0053]
In this embodiment, a trapezoidal membership function is used as the membership function constituting the template. For example, if the maximum amount of change in the X direction on a specific partial curve obtained from 5 pieces of X-sine data is “max” and the minimum value is “min”, the minimum value “min” is corrected. Divide by the value Wx (the correction value Wx is a value larger than 1; for example, 1.2) and multiply the maximum value “max” by the correction value Wx to correct the minimum value “min” and the maximum value “max”, respectively. As a membership function for the X-sine data of the partial curve, the width from the minimum value “min / Wx” after correction to the maximum value “max × Wx” is set as the width w, and the correction value Wy is set to the width w. A trapezoidal membership function as shown in FIG. 6 is generated with the added value (w + Wy). However, the correction values Wx and Wy are adjustment parameters for absorbing variations in positive sign.
[0054]
In addition, for the evaluation of the template generated in this way, the positive sign data excluding the five positive sign data used for creating the template out of the n sign data sampled by the processing of S120 to S140. And negative sign data is used. When the template is actually evaluated, the positive sign data and the negative sign data are input to the verification network as shown in FIG. The collation rate with the positive example sign data used to generate the template is obtained, and the correct collation rate ti obtained in the collation network is set to a value of 1 for positive example sign data and a value of 0 for negative example sign data. Thus, the fitness F is calculated using the following equation.
[0055]
[Expression 1]

[0056]
In the above equation, m1 and m2 represent the numbers of positive example sign data and negative example sign data used for template evaluation, respectively. In the above formula, the average of the matching errors for the positive sign and the negative sign is calculated as the fitness F, and the closer the fitness F is to 1, the more the generated template (in other words, the origin of the template). Therefore, in S420, the locus in the chromosome whose fitness F is closest to 1 is selected as an elite locus.
[0057]
The verification network shown in FIG. 7 represents the function of the signature verification unit 22 described above, and is a template composed of a plurality of membership functions for X signature data and Y signature data generated from a plurality of loci constituting a chromosome. And the degree of coincidence of each partial curve obtained from the X sine data and Y sine data for each partial curve corresponding to each membership function in the sine data to be collated with each membership function constituting this template And a multiplication unit for obtaining a matching rate between the signature data to be verified and the positive signature data used for creating the template. In other words, the same verification network is used for template evaluation as when sign verification is performed, and the positive and negative signature data and the positive signature data used to create the template are actually verified using the template to be evaluated. To make it happen. Then, using the generated membership function, the higher the degree of coincidence between the partial curves, the closer the value is to 1, and the matching rate obtained also approaches 1.
The series of processes from S380 to S420 includes m templates for m chromosomes each having m loci obtained by proliferating / mutating one gene in the chromosome selected in S370. It is a process for selecting elite loci from m loci by generating and evaluating each template, and these series of processes are m m each having m loci. It is possible to generate m templates for each chromosome at the same time and evaluate each template at the same time, or generate a template from the chromosome selected in S370 and evaluate this, and then in S380 Mutate any selected locus, generate a template from a chromosome with that locus, evaluate it, and mutate further Mutate the locus again, generate a template from the chromosome that has the locus, and evaluate this, then mutate one selected locus in sequence, and the template corresponding to that locus The elite loci may be selected by sequentially evaluating.
[0058]
When elite loci are selected as described above, it is determined whether generation update (ie selection of elite chromosomes) for all loci constituting the chromosome selected in S370 is completed (S430). If generation updates for all loci have not been completed, the process proceeds to S380. In S380, among the i loci constituting the chromosome currently targeted for genetic manipulation, an arbitrary genetic locus for which the elite locus has not been selected by the genetic manipulation is selected, and the subsequent processing after S390 is performed. Then, genetic manipulation is performed on the selected locus to select the elite locus. As a result, genetic manipulation is performed on all loci constituting the chromosome selected in S370, and elite loci are set for all the loci. If it is determined in S430 that the generation update of the locus has been completed, the process proceeds to S440, and the chromosome selected in S370 is updated to a chromosome composed of the selected elite locus.
[0059]
Next, when the chromosome is updated in S440, it is determined whether or not all k chromosomes have been updated by the processing of S370 to S440 (S450). And if the update with respect to all the chromosomes is not completed, it will transfer to S370 again. Then, in S370, an arbitrary chromosome that has not been updated is selected from the k chromosomes, and in the subsequent processes after S380, the genetic manipulation and the elite locus of each locus constituting the selected chromosome are performed. Selection is made and this chromosome is updated. As a result, generational changes are made for all k chromosomes generated by the processing of S310 to S360.
[0060]
Next, when it is determined in S450 that all k chromosomes have been updated, the process proceeds to S460, and the k chromosomes updated as described above are ordered in ascending order of template evaluation. A process for so-called chromosomal selection that deletes a predetermined percentage (for example, 20%) is executed. In addition, for selection of the chromosome to be deleted, for example, the evaluation result (that is, fitness F) obtained when the k chromosomes are changed generations by the processes of S370 to S440 can be used.
[0061]
Then, when a predetermined percentage of the low-evaluated chromosomes are deleted from the k chromosomes as described above, any two chromosomes are selected from the remaining chromosomes in S470, and one chromosome is constructed. Chromosomes deleted in S460 by performing a so-called chromosome crossover process in which a new chromosome is generated by combining loci with a predetermined stroke number or higher and loci with a lower stroke number constituting the other chromosome. Replenish new chromosomes as many as
[0062]
When chromosomes are supplemented in this way, the process proceeds to S480, a template is generated for each supplemented chromosome, and the generated template is evaluated in subsequent S490. Note that the template generation and its evaluation are executed in the same procedure as in S400 and S410.
[0063]
In the subsequent S500, based on the evaluation result of the new chromosome obtained by the process of S490 and the evaluation result of the remaining chromosomes obtained by the process of S370 to S440, the most evaluated out of the k chromosomes. Are selected as elite chromosomes (that is, fitness F is closest to 1). In subsequent S510, it is determined whether or not the fitness F of the selected elite chromosome is equal to or higher than a predetermined criterion (for example, 95%), and the fitness F of the elite chromosome is equal to or higher than the criterion. If there is, the elite chromosome optimally reflects the personality characteristics of the sign registrant, and if the template generated based on this is used, it is determined that signature verification can be performed without any problem, and the gene calculation process is terminated. .
[0064]
On the other hand, if it is determined in S510 that the fitness F of the elite chromosome has not reached the criterion, the process proceeds to S370 to generate an elite chromosome having a fitness equal to or higher than the criterion by performing genetic manipulation again. To do. As a result, in the processes after S370, the genetic manipulation is again performed on the k second-generation chromosomes consisting of the chromosome updated by the genetic manipulation and the newly generated chromosome by the selection and crossover. Eventually, an optimal elite chromosome having a fitness F equal to or higher than the criterion is selected, and the gene calculation process is completed.
[0065]
The gene operation shown in FIG. 3 is substantially the same as that described in the above-mentioned paper, and the technology disclosed in this paper and this example are the structure of the gene locus constituting the chromosome, and this The structure of the template generated from the gene locus is different. And the procedure of gene calculation is not limited to the procedure of the flowchart shown in FIG. 3, It can change suitably.
[0066]
Next, returning to FIG. 2, when an elite chromosome having a fitness F equal to or higher than the criterion is selected by the gene calculation process in S150, a signature verification template is generated from the elite chromosome in S160.
In S160, a template for signature verification may be generated from the elite chromosome. However, since the template corresponding to the elite chromosome has already been generated at the time of executing the genetic operation processing in S150, the template is read. It may be. In addition to the membership function for each partial curve, the template generated in S160 also includes chromosome information consisting of a plurality of loci in order to identify the partial curve corresponding to the membership function. It is.
[0067]
Then, when the signature verification template corresponding to the elite chromosome is generated in this manner, the process proceeds to S170, and a message for causing the sign registrant to input the ID number is displayed on the display device 6, An ID code input process is executed to accept that the sign registrant operates the keyboard 4 to input an ID number.
[0068]
In the subsequent S180, the ID number of the sign registrant captured in the ID code input process in S170 is assigned to the signature verification template generated or read in S160 and stored in the template storage medium 20, and the process is temporarily performed. finish.
[0069]
The above calculation processing at the time of sign registration is not uniform, but is completed in tens of minutes to a few minutes or less. The data size of the template generated as described above usually falls within a capacity of about 500 bytes if one membership function is composed of 16 bytes.
[0070]
Next, calculation processing executed by the microcomputer 10 at the time of signature verification will be described.
As described above, the microcomputer 10 determines whether or not the signature is registered in S110. If not, the microcomputer 10 proceeds to S210. In S210, a message for causing the person to be collated to input his / her ID number is displayed on the display device 6, and accepting that the person to be collated operates the keyboard 4 to input the ID number is input. Execute the process.
[0071]
In S210, when the ID number of the person to be verified is taken in, a template corresponding to the ID number is called from the template storage medium 20 in S220, and in S230, the person to be verified is signed on the display device 6. A signature input process for reading signature data input from the tablet 2 when the person to be verified writes a signature on the tablet 2 is executed.
[0072]
Then, when the signature data of the person to be collated is obtained by this signature input process, the process proceeds to S240, and the signature data is normalized. In S250, the normalized signature data and template are used. A matching operation is performed to match the signature input by handwriting with the positive signature used to generate the template. Note that, as described above with reference to FIG. 7, this matching operation takes in partial curve data corresponding to the membership function constituting the template from the sign data normalized in S240, and the data Fuzzy inference using the corresponding membership function and the degree of coincidence between the template and the data at the time of template creation, and then by multiplying the degree of coincidence of the data for each partial curve obtained by this fuzzy inference, registration with handwritten signature It is executed by a procedure such as obtaining the verification rate with the signature.
[0073]
When the matching rate is obtained by the matching operation in S250 in this way, in the subsequent S260, the matching rate is equal to or higher than a determination value (for example, 0.7) that can authenticate the person to be verified as the signature registrant. If the verification rate is equal to or higher than the determination value, the verification target person is authenticated as the signature registrant, and conversely, if the verification rate does not reach the predetermined value, the verification target person is signed. Is entered again, or a collation determination process is executed to determine whether the person to be collated is a fake person. In S270, the result of the collation determination process is output to the external device, and the process is temporarily terminated.
[0074]
As described above, according to the signature verification apparatus of the present embodiment, in accordance with the genetic algorithm, the partial curve that most reflects the personality characteristics of the registrant is extracted from the positive sign described by the registrant, A template for signature verification is generated by generating a membership function for fuzzy inference of the degree of coincidence for each partial curve. The gene locus for specifying the partial curve is composed of the number of the stroke to which the partial curve belongs and three pieces of information indicating the start point and end point of the partial curve on the stroke. Corresponding to the seat, the position change amount of the sign in the X direction and the Y direction corresponding to the length of the partial curve.
[0075]
Therefore, according to the signature verification apparatus of the present embodiment, at the time of signature verification, it is possible to fuzzy infer the degree of coincidence between the registered signature and the handwritten signature using a partial curve that most reflects the personality characteristics of the registered signature. The accuracy of signature verification can be ensured, and the capacity of the template used for signature verification (in other words, the number of membership functions) can be reduced to reduce the capacity of the template storage medium 20 for storing it. Can do.
[0076]
As mentioned above, although one Example of this invention was described, this invention is not limited to the said Example, It can take a various aspect.
For example, in the above embodiment, signature registration and signature verification are realized by a single signature verification device. However, at the time of signature registration, a dedicated computer is used to perform high-speed arithmetic processing according to a genetic algorithm. At the time of signature verification, the signature verification may be performed by incorporating a storage medium registered in the signature verification terminal device or the like.
[0077]
Moreover, in the said Example, although the number of the gene loci which comprises a chromosome was demonstrated as what is suitably set using a random number etc. within the range of 5-50, in this case, the elite chromosome finally obtained The number of gene loci is also set as appropriate, and the data size of the template generated from this elite chromosome changes each time. For this reason, for example, when it is desired to keep the template data size constant, the number of loci is added to one of the evaluation items when selecting an elite chromosome, and the number of loci is a predetermined value as an elite chromosome. What is necessary is just to make it comprise so that a chromosome may be selected or to fix the number of gene loci which comprise a chromosome to a predetermined value. In this way, the data size of the template can always be made constant.
[0078]
In the above-described embodiment, the negative example signature data described by a person other than the signature registrant is sampled for the evaluation of the generated template. However, for example, a filter that generates random numbers and limits the amplitude. The negative example signature data may be generated from the positive example signature data by a procedure such as generating negative example signature data by multiplying the positive example signature data by multiplying the first example signature data.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration and functions of an entire signature verification apparatus according to an embodiment.
FIG. 2 is a flowchart showing a calculation process executed at the time of signature registration and signature verification by the microcomputer constituting the signature verification apparatus of the embodiment.
FIG. 3 is a flowchart showing details of gene operation processing executed in S150 of FIG.
FIG. 4 is an explanatory diagram illustrating an example of time-series data input from a tablet.
FIG. 5 is an explanatory diagram for explaining gene loci and chromosomes generated based on time-series data input from a tablet during signature registration.
FIG. 6 is a diagram illustrating a procedure for generating a membership function for fuzzy inference of the degree of coincidence of partial curves specified by a gene locus.
FIG. 7 is an explanatory diagram illustrating the configuration of a template generated at the time of sign registration and the function of a verification network that performs signature verification using the template.
[Explanation of symbols]
2 ... Tablet 4 ... Keyboard 6 ... Display (LCD)
DESCRIPTION OF SYMBOLS 10 ... Microcomputer 12 ... Signal processing part 14 ... Gene operation part
16 ... Template generation unit 20 ... Template storage medium
22 ... Signal verification unit 24 ... Determination output unit 28 ... Control unit

Claims (7)

Time series data consisting of the position and writing pressure at the time of writing the signature is taken, the time series data is normalized by the size and position of the signature, and a plurality of parts representing the personality characteristics of the signature from the normalized time series data Set a curve, generate a membership function for fuzzy inference of the degree of coincidence with the signature data to be verified for each partial curve, and use the generated membership function for each partial curve for sign verification A method for creating a signature verification template to be set as a template for
When setting the partial curve,
First, by selecting an arbitrary stroke from a plurality of strokes obtained by dividing the normalized time series data into regions where signs are continuous, and selecting an arbitrary section on the selected stroke as the partial curve. The plurality of partial curves are initialized, and the plurality of partial curves are respectively set to a stroke number to which the partial curve belongs, a start point of the partial curve on the stroke, and an end point of the partial curve on the stroke. Set as a genetic locus consisting of three types of information
Next, using the locus group consisting of a plurality of loci representing each partial curve as the chromosome of the signature, by operating the locus according to a preset genetic algorithm, the personality characteristics of the signature are optimized. Generating an elite chromosome to be represented, and setting a partial curve specified at each gene locus constituting the generated elite chromosome as the partial curve for creating the template,
A method for creating a template for signature verification characterized by The genetic algorithm propagates and mutates arbitrary loci constituting the chromosome to generate a plurality of chromosomes having different loci, and optimally represents the personality characteristics of the signature among the plurality of chromosomes. By selecting the chromosome, the operation of extracting the most suitable locus for expressing the individuality characteristics of the signature is performed for each locus constituting the chromosome, thereby generating the elite chromosome. 2. The method for creating a signature verification template according to claim 1, wherein the signature verification template is an improved genetic algorithm. As a membership function for each partial curve, the time series data at the time of sign writing is captured multiple times, and the minimum and maximum values of the data change amount from the start point to the end point of each partial curve in the plurality of time series data The method for generating a signature verification template according to claim 1 or 2, wherein a trapezoidal membership function having a width from the minimum value to the maximum value is generated. A sign input means for inputting time-series data corresponding to a handwritten signature by sequentially detecting the position and writing pressure at the time of writing the signature;
Normalization means for normalizing the time-series data at the time of sign writing input from the sign input means by the size and position of the sign;
A partial curve setting means for setting a plurality of partial curves representing individual characteristics of the signature from the time series data normalized by the normalization means;
For each partial curve set by the partial curve setting means, a membership function for fuzzy inference of the degree of coincidence with the signature data to be verified is generated, and the membership function for each generated partial curve A template generation means for storing a template for sign verification in a predetermined template storage medium;
A signature verification template creation device comprising:
The partial curve setting means includes
By selecting an arbitrary stroke from a plurality of strokes obtained by dividing the normalized time series data for each region where signs are continuous, and selecting an arbitrary section on the selected stroke as the partial curve, Initializing a plurality of partial curves, each of the plurality of partial curves represents a stroke number to which the partial curve belongs, a start point of the partial curve on the stroke, and an end point of the partial curve on the stroke 3 A partial curve initial setting means for setting as a gene locus comprising species information;
A locus group consisting of a plurality of loci set by the partial curve initial setting means is used as a chromosome of the signature input by handwriting, and the locus of the signature is manipulated according to a preset genetic algorithm. An elite chromosome generating means for generating elite chromosomes that optimally represent individuality characteristics, and setting a partial curve specified at each locus constituting the generated elite chromosome as a partial curve for creating the template;
An apparatus for creating a template for signature verification, comprising: The elite chromosome generating means includes
By multiplying and mutating arbitrary loci constituting the chromosome to generate a plurality of chromosomes having different loci, and selecting a chromosome that optimally represents the personality characteristics of the signature from the plurality of chromosomes Generating the elite chromosome according to a locally improved genetic algorithm, wherein an operation for extracting the locus most suitable for representing the individuality characteristic of the signature is performed for each locus constituting the chromosome. The signature verification template creation apparatus according to claim 4. The template generation means fetches the time series data at the time of sign writing a plurality of times via the sign input means, and the minimum value and the maximum data change amount from the start point to the end point of each partial curve in the plurality of time series data. 6. The signature verification template creation apparatus according to claim 4, wherein a trapezoidal membership function having a width from the minimum value to the maximum value is generated by obtaining a value. A template storage medium in which a template created by the creation device according to claim 4 is stored in advance;
A sign input means for inputting time-series data corresponding to a handwritten signature by sequentially detecting the position and writing pressure at the time of writing the signature;
Normalization means for normalizing the time-series data at the time of sign writing input from the sign input means by the size and position of the sign;
Partial curve data corresponding to a plurality of membership functions stored as templates in the template storage medium is extracted from the time-series data normalized by the data normalization means, and the data corresponds to the data. An inference means for fuzzy inferring a degree of coincidence for each partial curve between the registered signature and the handwritten signature from a membership function;
A collation rate output means for calculating the total coincidence between the registered signature and the handwritten signature from the coincidence for each partial curve fuzzy inferred by the inference means, and outputting the calculation result as a collation rate;
A signature verification apparatus comprising:

Images