High precision on-line signature dynamic verification system

• US 5,202,930 A
• Filed: 04/09/1991
• Issued: 04/13/1993
• Est. Priority Date: 09/05/1989
• Status: Expired due to Term

• Current Assignee: Verifax Corp.
• Original Assignee: Alexander Livshitz, Ruth Shraiman
• Inventors: Livshitz, Alexander, Shraiman, Ruth
• Primary Examiner(s): Moore, David K.
• Assistant Examiner(s): CAMMARATA, MICHAEL
• Application Number:
  US07/683,533
• Time in Patent Office: 735 Days
• Field of Search: 382/2, 382/3, 382/13, 382/18, 382/51, 382/20, 382/41, 382/49
• US Class Current: 382/122
• CPC Class Codes: G06V 40/30 Writer recognition; Reading and verifying signatures
• CPC Class Codes: G06V 40/37 based only on signature signals such as velocity or pressure, e.g. dynamic signature recognition
• CPC Class Codes: G07C 9/35 by means of a handwritten signature

Abstract

On-line handwritten data verification, such as a signature, is obtained based on the comparison of the dynamics of a sample and a reference signatures by use of a new implementation of correlation function analysis. The dynamic data concerning scriber movement during the process of signature making is recorded and converted to digital form. Prior to correlation function analysis for evaluating a to-be-verified signature, reference signature dynamic data and to-be-verified signature dynamic data digital signals are pre-processed to eliminate time distortions. The compared signals are reduced to the same time scale, thereby providing a frequency coincidence between the signals. In order to eliminate the phase distortions, a "sliding window" mechanism is used to establish the mapping between phase coincident regions of the reference and of the to-be-verified signals. Then the multi-dimensional cross-correlation function analysis is applied to the pair of indivisible stationary signals. The sliding window mechanism provides two additional criteria to distinguish between authentic and forged signatures by determination of different characteristics of time distortions for such signals. The resulting measures of similarity are then compared with thresholds that have been selected to determine acceptance or rejection of the signature. The method and the apparatus can be applied for signature authentication in a wide variety of applications like security of physical access, computer network access, facsimile legalization, credit card industry, and many others.

Claims

1. A machine method of on-line handwritten signature verification based on the dynamics of a reference signature and of a to-be-verified signature in accordance with one or more selected threshold value, including the steps of:

a. gathering a set of reference analog dynamic signals related to scriber movement for handwritten reference signatures during the process of reference signature making, such signals being selected from the group consisting of forces transmitted from the hand of the writer to the scriber, scriber acceleration, scriber velocity, and mixtures thereof, all as functions of time;

b. converting the reference analog dynamic signals from analog to digital form, and storing the resulting reference digital signals in a file of reference digital signals for retrieval therefrom;

c. gathering a set of to-be-verified analog dynamic signals related to scriber movement for a to-be-verified handwritten signature during the process of to-be-verified signature making, such signals to be selected from the group consisting of forces transmitted from the hand of the writer to the scriber, scriber acceleration, scriber velocity, and mixtures thereof, all as functions of time;

d. converting the to-be-verified analog dynamic signature signals from analog to digital form;

e. providing a pair of digital signals, said digital signal pair being composed of the to-be-verified digital signals and one of the reference digital signals taken from the reference signal file, with which it is to be compared;

f. processing the data pair of digital signals to eliminate time distortions between them utilizing the steps of;

i. using a sliding window to find phase shift locations, and thereby establishing the mapping between phase coincident areas of a reference digital signal and of the to-be-verified digital signals;

ii. eliminating phase shift distortions between the reference digital signal and of the to-be-verified digital signal by the use of the mapping procedure of step f. ii.;

g. determining the similarity of a pair of reference and to-be-verified signatures utilizing the steps of;

i. evaluating the cross-correlation matrix K_rs from step f. ii., with each element K_ij representing the maximum value of the corresponding cross-correlation function for each pair of x, y and z components of digital signal vector;

##EQU4## ii. utilizing information in said matrix to produce a measure, said measure being determined using a method selected from the method of calculating a trace of the matrix and the method of calculating a norm of the matrix;

iii. using the measure as determined by step g.ii., and comparing it with an appropriate threshold in order to produce a non-rejection or rejection message of the authenticity of to-be-verified signature, as a criterion Cr₁ ;

iv. determining distribution of phase distortion for windows by constructing a histogram of the first order shift differences obtained from the application of the sliding window method in step f.i.;

v. utilizing the information of said first order shift differences histogram to produce a measure which characterizes the portion of the first order shift differences in the zero neighborhood;

vi. using the measure determined in step g.v. and comparing it with an appropriate threshold in order to produce a message of non-rejection or rejection of authenticity of the to-be-verified signature, as a criterion Cr₂ ;

vii. determining coincidence of phase distortion distributions for two of the different components of the signature signal vectors;

X, Y and Z by constructing histogram of the shift differences, obtained from application of window method in step f.i., for each component by comparing their shift histograms;

viii. utilizing the information of said shift differences histogram to produce a measure which characterizes the portion of the shift differences for the two considered vector components of the signature signal which values are in the zero neighborhood;

ix. using the measure determined in step g. viii. and comparing it with an appropriate threshold in order to produce a signal of non-rejection or rejection of authenticity of the to-be-verified signature, as criterion Cr₃ ; and

then

x. computing the verification result using criterion selected from the group consisting of Cr₁, Cr₂, and Cr₃, and combinations thereof;

xi. using the verification result of g.x. to produce a signal of approval or rejection of the authenticity of to-be-verified signature.

2. The method of claim 1 wherein the correlation analysis utilizes a sub-matrix of g.i.

3. The method of claim 1 wherein the information in said matrix is utilized in step g.ii. to determine a trace of the matrix.

4. The method of claim 1 wherein the information in said matrix is utilized in step g.ii. to determine a norm of the matrix.

5. The method of claim 1 wherein determination of the similarity of signatures is based on evaluation of cross-correlation function for a selected component of the signature signals.

6. The method of claim 5 wherein correlation analysis is utilized by evaluating cross-correlation function for the magnitude of the selected components of the signature signals.

7. The method of claim 1, wherein the sliding window method in step f. i. is used to find the position of maximum correlation functions between windows of one signal and the corresponding pieces of the other signal for determination of phase distortion locations.

8. The method of claim 1 wherein the sliding window in step f.i. is a piece of one signal called window which is moved along the signal with a predetermined step of a size which is less than size of the window, and which is used to find for each window the corresponding piece in another signal of the pair on the basis of the position of maximum value of cross-correlation function between them.

10. The method of claim 1 wherein the verification result of step g.x. is computed by determining the conjunction of measures Cr₁, Cr₂, and Cr₃.

11. The method of claim 1 wherein prior to step f.i. the digital signals of the data pair are reduced to the same average velocity to achieve substantial frequency coincidence between signals in the pair.

9. Apparatus for use in on-line handwritten data verification based on the dynamics of a reference and of a to-be-verified signature signals in accordance with one or more selected threshold values, including in combination:

a. a scriber;

b. means for gathering a set of reference analog dynamic signals related to scriber movement for reference during a process of reference signature making, and analog dynamic signals during a process of to-be-verified signature making, such signals being selected from the group consisting of forces transmitted from the hand of the writer to the scriber, scriber acceleration, scriber velocity, and mixtures thereof, all as functions of time;

c. means for converting the analog reference and to-be-verified dynamic signals from analog to digital form;

d. means for storing the resulting reference digital signals in a file of reference signature signals;

e. means for providing a data pair composed of the digital data of the to-be-verified signal and the digital data of one of the reference file signals with which it is to be compared;

f. means for processing the data pair of digital signals to eliminate time distortions between them utilizing the steps of;

i. using a sliding window method to find phase shift locations, and thereby establishing the mapping between phase coincident areas of a reference digital signal and of the to-be-verified digital signals;

ii. eliminating phase shift distortions between the reference digital signal and of the to-be-verified digital signal by the use of the mapping procedure of step f. i.;

g. means for determining the similarity of a pair of reference and to-be-verified signatures utilizing the steps of;

i. evaluating the cross-correlation matrix from step f. ii. as a numerical matrix with each element representing the maximum value of the corresponding cross-correlation function K_rs for each pair of x, y and z components of digital signal vectors;

##EQU5## ii. utilizing information in said matrix to produce a measure, said measure being determined using a method selected from the method of calculating a trace of the matrix and the method of calculating a norm of the matrix;

iii. using the measure as determined by step g.ii., and comparing it with an appropriate threshold in order to produce a non-rejection or rejection signal of the authenticity of to-be-verified signature, as a criterion Cr₁ ;

iv. determining distribution of phase distortion for windows by constructing a histogram of the first order shift differences obtained from the application of the sliding window method in step f.i.;

v. utilizing the information of said first order shift differences histogram to produce a measure which characterizes the portion of the first order shift differences in the zero neighborhood;

vi. using the measure determined in step g.v. and comparing it with an appropriate threshold in order to produce a message of non-rejection or rejection of authenticity of the to-be-verified signature, as a criterion Cr₂ ;

vii. determining coincidence of phase distortion distributions for two components of the signature signal vectors;

X, Y and Z by constructing histogram of the shift differences;

viii. utilizing the information of said shift differences histogram to produce a measure which characterizes the portion of the shift differences for the two considered vector components of the signature signals which values are in the zero neighborhood;

ix. using the measure determined in step g. viii. and comparing it with an appropriate threshold in order to produce a signal of non-rejection or rejection of authenticity of the to-be-verified signature as criterion Cr₃ ; and

then

x. means for computing the verification result using criterion selected from the group consisting of Cr₁, Cr₂, and Cr₃, and combinations thereof;

xi. means for using the verification result of g.x. to produce a signal of approval or rejection of the authenticity of to-be-verified signature.

12. A machine method of on-line handwritten signature verification using on the dynamics of scriber movement while recording a reference signature and a to-be-verified signature in accordance with one or more selected threshold value, including the steps of:

a. gathering a set of analog dynamic reference signals related to movement of the scriber for handwritten reference signatures during the process of reference signature making, such signals being selected from the group consisting of forces transmitted from the hand of the writer to the scriber, scriber acceleration, scriber velocity, and mixtures thereof, all said being measured as functions of time;

b. converting the reference analog dynamic signals from analog to digital form, and storing the resulting reference digital signals in a file of reference digital signals for retrieval therefrom;

c. gathering a set of to-be-verified analog dynamic data signals related to movement of said scriber for a to-be-verified handwritten signature during the process of to-be-verified signature making, such signals to be selected from the group consisting of forces transmitted from the hand of the writer to the scriber, scriber acceleration, scriber velocity, and mixtures thereof, all as functions of time;

d. converting the to-be-verified analog dynamic signature signals to digital form;

e. providing a pair of digital signals, said digital signal pair being composed of the to-be-verified digital signal and one of the reference digital signals taken from the reference signal file;

f. processing the pair of digital signals to eliminate time distortions between them utilizing the steps of;

i. using a sliding window to find phase shift locations, and thereby establishing the mapping between phase coincident areas of reference digital signal and of the to-be-verified digital signal;

ii. eliminating phase shift distortions between the reference digital signal and the to-be-verified digital signal by the use of the mapping procedure of step f. i.;

g. determining the similarity of the pair of reference and to-be-verified signatures using criterion selected from the group consisting of Cr₁, Cr₂, and Cr₃, and combinations thereof;

wherein Cr₁ is determined by;

i. evaluating the cross-correlation matrix K_rs based on data from step f. ii., with each element representing the maximum value of the corresponding cross-correlation function for each pair of x, y and z components of digital signal vectors;

##EQU6## ii. producing a measure from the matrix, said matrix being created using a method selected from the method of calculating a trace of the matrix and the method of calculating a norm of the matrix;

iii. comparing the measure as determined by step g.ii., with an appropriate threshold in order to produce a non-rejection or rejection signal of the authenticity of the to-be-verified signature, as a criterion Cr1;

wherein Cr₂ is determined by;

i. determining distribution of phase distortion for windows by constructing a histogram of the first order shift differences obtained from the application of the sliding window method in step f. i;

ii. utilizing the information of said first order shift differences histogram to produce a measure which characterizes the portion of the first order shift differences in the zero neighborhood;

iii. using the measure determined in previous step ii and comparing it with an appropriate threshold in order to produce a signal of non-rejection or rejection of authenticity of the to-be-verified signature, as a criterion Cr₂ ;

wherein Cr₃ is determined by;

i. determining coincidence of phase distortion distributions for two components of the signature signal vectors;

X, Y and Z by constructing histogram of the shift differences obtained from the application of the sliding window method in step f.i;

ii. utilizing the information of said shift differences histogram to produce a measure which characterizes the portion of the shift differences for the two considered vector components of the signature signal which values are in the zero neighborhood;

iii. comparing the measure determined in previous step ii. with an appropriate threshold in order to produce a signal of non-rejection or rejection of authenticity of the to-be-verified signature as criterion Cr₃ ; and

then

h. using the determination of similarity from g., to produce a signal of approval or rejection of the authenticity of to-be-verified signature.

13. Apparatus for use in on-line handwritten data verification using the dynamics of the scriber movement while recording a reference and a to-be-verified signature signals in accordance with one or more selected threshold values, including in combination:

a. a scriber;

b. means for gathering a set of reference analog dynamic signals related to movement of the scriber during the process of reference signature making and during a process of to-be-verified signature making, such signals being selected from the group consisting of forces transmitted from the hand of the writer to the scriber, scriber acceleration, scriber velocity, and mixtures thereof, all as functions of time;

c. means for converting the analog reference and to-be-verified dynamic signals from analog to digital form;

d. means for storing the resulting reference digital signals in a file of reference digital signals;

e. means for providing a data pair composed of the digital data of the to-be-verified signal and the digital data of one of the reference file signals;

f. means for processing the data pair of digital signals to eliminate time distortions between them utilizing the steps of;

i. using a sliding window to find phase shift locations, and thereby establishing the mapping between phase coincident areas of a reference digital signal and of the to-be-verified digital signal;

ii. eliminating phase shift distortions between the reference digital signal and the to-be-verified digital signal by the use of the mapping procedure of step f.i.;

g. means for determining the similarity of the pair of reference and to-be-verified signatures using criterion selected from the group consisting of Cr₁, Cr₂, and Cr₃, and combinations thereof;

wherein Cr₁ is determined by;

i. evaluating the cross-correlation matrix K_rs based on data from step f. ii., with each element representing the maximum value of the corresponding cross-correlation function for each pair of x, y and z components of digital signal vectors;

##EQU7## ii. producing a measure from the matrix, said measure being created using a method selected from the method of calculating a trace of the matrix and the method of calculating a norm of the matrix;

iii. comparing the measure as determined by step g.ii., with an appropriate threshold in order to produce a non-rejection or rejection signal of the authenticity of the to-be-verified signature, as a criterion Cr1;

wherein Cr₂ is determined by;

i. determining distribution of phase distortion for windows by constructing a histogram of the first order shift differences obtained from the application of the sliding window method in step f. i;

ii. utilizing the information of said first order shift differences histogram to produce a measure which characterizes the portion of the first order shift differences in the zero neighborhood;

iii. comparing the measure determined in previous step ii. with an appropriate threshold in order to produce a signal of non-rejection or rejection of authenticity of the to-be-verified signature as a criterion Cr₂ ;

wherein Cr₃ is determined by;

i. determining coincidence of phase distortion distributions for two components of the signature signal vectors;

X, Y and Z by constructing histogram of the shift differences;

ii. utilizing the information of said shift differences histogram to produce a measure which characterizes the portion of the shift differences for the two considered vector components of the signature signal which values are in the zero neighborhood;

iii. comparing the measure determined in previous step ii. with an appropriate threshold in order to produce a signal of non-rejection or rejection of authenticity of the to-be-verified signature as criterion Cr₃ ; and

h. means for using the determination of similarity from g., to produce a signal of approval or rejection of the authenticity of a to-be-verified signature.