JP2005149353A - Fingerprint collation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a computer system or the like capable of performing rapid and precise fingerprint collation with a minimized work memory. <P>SOLUTION: This computer system for fingerprint collation comprises an input means for inputting information on characteristic points of a fingerprint; a fingerprint information storage means; a similarity calculation means; a first position/rotation calculation means; a second position/rotation calculation means; a determination means for correcting the difference in position and rotation between a registered fingerprint calculated by the second position/rotation calculation means and an input fingerprint, calculating the similarity by the similarity calculation means in the corrected position and rotation, and determining whether the owner of the registered fingerprint is identical with the owner of the input fingerprint according to the calculated similarity; and an output means for outputting a determination result by the determination means. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a computer system for fingerprint collation and the like. More specifically, when calculating the difference between the registered fingerprint and the difference in the amount of movement and the amount of rotation of the input fingerprint, first determine the approximate values, and then determine the detailed difference in the amount of movement and the difference in the amount of rotation. The present invention relates to a computer system for epoch-making fingerprint collation that achieves a small memory.

  The fingerprint collation system is a system for identifying an individual using the fact that human fingerprints are different. More specifically, the fingerprint collation system reads a collator's fingerprint information with a fingerprint reading unit such as a sensor, and performs personal recognition by collating the fingerprint information that has been registered and processed in advance with the read fingerprint information. System.

  FIG. 1 is a diagram illustrating an example of a fingerprint collation system. As shown in FIG. 1, a fingerprint collation system 1 includes a fingerprint sensor 2, a preprocessing unit 3, a feature extraction unit 4, a database unit 5 that stores a plurality of templates 6, a collation unit 7, an external unit And a display unit 8.

  Hereinafter, an example of fingerprint collation will be described with reference to FIG. Fingerprint verification consists of two steps: registration and verification. When registering fingerprint information, the fingerprint sensor 2 is caused to read the fingerprint. The fingerprint information read by the fingerprint sensor 2 is subjected to predetermined preprocessing in the preprocessing unit 3 before the feature points are extracted. For example, fingerprint information is converted into digital information by an A / D conversion circuit, noise is reduced by a filter circuit, binarized by a binarization circuit, and thinned by a thinning circuit. The feature extraction unit 4 extracts feature information that is information about feature points such as the branch point 9 and the end point 10 from the pre-processed fingerprint information. The extracted feature information is stored in the database unit 5 in the fingerprint matching system 1 as a template (Tem).

  The feature points described above can be understood with reference to FIG. 2A is a diagram showing a part of a fingerprint having a branch point 9 and FIG. 2B a feature point of which is an end point 10. FIG. 2C is a photograph replacing a drawing showing a fingerprint including feature points. As shown in FIG. 2C, the fingerprint is composed of raised lines flowing in various patterns called ridges and valleys between the ridges. The branch point 9 is a feature point where the ridge is divided from one path into two paths at the Y junction. On the other hand, the end point 10 is a feature point where the ridge ends.

  When collating a fingerprint, the fingerprint sensor 2 is caused to read the fingerprint in the same manner as the registration process described above. The fingerprint information read by the fingerprint sensor 2 is subjected to predetermined preprocessing in the preprocessing unit 3 before the feature points are extracted. For example, fingerprint information is converted into digital information by an A / D conversion circuit, noise is reduced by a filter circuit, binarized by a binarization circuit, and thinned by a thinning circuit. The feature extraction unit 4 extracts feature information that is information about feature points such as the branch point 9 and the end point 10 from the pre-processed fingerprint information.

  The collation unit 7 collates the fingerprint feature information (In) input for collation with the fingerprint feature information (Tem) registered in advance, and outputs a collation result. There are various algorithms for the collation unit 7 to collate feature information, and they are devised according to needs ("Personal authentication system using fingerprints" AK Jain, L. Hong, S. Pakaniti). Proceedings of the IEEE, Vol. 85, No. 9, pages 1365 to 1388, 1977 (Non-Patent Document 1), “Online fingerprint verification” AK Jain, L. Hong and R. Boll; IEEE Transactions on PAMI, Vol. 19, No. 4, pp. 302-314, 1977 (Non-patent Document 2), “Computer Algebra for Fingerprint Verification”, S Bistarelli, G Boffi, F. Rossi; ICCS Workshop on Computer Alegebra System and ther Applications, 2003 (Non-Patent Document 3), [8] “Feature Point Ridge Shape Algorithm for High-Speed Online Fingerprint Identification System” Mohamed Mostafa, Dongju Li and Hiroaki Kunieda; ISPACS 2000 Proceeding (Non-Patent Document 4) checking.).

  In the fingerprint registration process and the collation process, the position and direction of the finger are deviated. Therefore, it is necessary to perform collation after correcting this. Further, since the pressure for pressing the finger against the fingerprint sensor 2 is different between the fingerprint registration process and the collation process, scaling that the size of the fingerprint image is different occurs. Since the fingerprint registration process and the collation process are different in timing, there may be a situation in which a part of the fingerprint is cracked due to an accident or injury after registering the fingerprint. In the fingerprint registration process and the collation process, a change may occur in the fingerprint image read by the fingerprint sensor 2 because the collator is sweating or the collider's finger is dirty.

  A computer system consisting of various circuits has been developed for the conventional fingerprint verification system, but in order to accurately verify fingerprint feature information, various corrections must be made in consideration of the above circumstances. did not become. For this reason, there is a problem that a large working memory is required for the computer system.

“Personal authentication system using fingerprint” K. Jain, L.M. Hong, S .; Pakaniti and R.A. Boll; Proceedings of the IEEE, Vol. 85, No. 9, pages 1365 to 1388, 1977 “Online fingerprint verification” K. Jain, L.M. Hong and R.K. Boll; IEEE Transactions on PAMI, Vol. 19, No. 4, pp. 302-314, 1977 “Computer Algebra for Fingerprint Verification” S Bistarelli, G Boffi, F.M. Rossi; ICCS Works on Computer Alegebra System and ther Applications, 2003 “Feature Point Ridge Shape Algorithm for High Speed Online Fingerprint Identification System” Mohamed Mostafa, Dongju Li and Hiroaki Kunieda; ISPACS 2000 Proceeding

  It is an object of the present invention to provide a computer system for fingerprint verification that takes into account the difference between the fingerprint registration process and the fingerprint verification process.

  An object of the present invention is to provide a computer system for fingerprint matching based on an unprecedented algorithm.

  An object of the present invention is to provide a computer system for fingerprint collation that can achieve highly accurate fingerprint collation with less memory than conventional ones.

  An object of the present invention is to provide a computer system for fingerprint collation that can achieve high-precision fingerprint collation at a higher speed than conventional ones.

  An object of the present invention is to provide a fingerprint collation system that can collate fingerprints quickly and accurately and requires less work memory.

  [1] In order to solve the above problems, a computer system according to the present invention includes an input unit that inputs information about a feature point of a fingerprint, information about a feature point of a registered fingerprint that is input by the input unit, and the input unit. Fingerprint information storage means for storing information about the feature points of the input fingerprint inputted by the above, similarity calculation means for obtaining the similarity of two fingerprints from the information about the feature points of the two fingerprints inputted, and the similarity First position calculating means for obtaining an approximate value of the difference between the position of the registered fingerprint stored in the fingerprint information storing means and the position of the input fingerprint using the degree calculating means, and the registration calculated by the first position calculating means After correcting the difference between the position of the fingerprint and the position of the input fingerprint, first rotation calculation means for roughly determining the difference in rotation between the registered fingerprint and the input fingerprint using the similarity calculation means. A first position / rotation calculating means for obtaining a rough position difference between the registered fingerprint and the input fingerprint, and a rotation difference, and the registered fingerprint and the input fingerprint calculated by the first position / rotation calculating means; Using the approximate value of the difference in position and the difference in rotation and the similarity calculation means, the difference between the position of the registered fingerprint stored in the fingerprint information storage means and the position of the input fingerprint is calculated in the vicinity of the approximate value. Second position calculating means for obtaining details in a narrow range compared to one position calculating means, information on the difference between the registered fingerprint position calculated by the second position calculating means and the input fingerprint position, and the similarity Using the calculation means, the difference between the rotation of the registered fingerprint stored in the fingerprint information storage means and the rotation of the input fingerprint is calculated in detail in a range narrower than the first rotation calculation means in the vicinity of the approximate value. Second rotation calculation A second position / rotation calculation means for obtaining a detailed value of a difference in position and a difference in rotation between the registered fingerprint and the input fingerprint, and a registered fingerprint calculated by the second position / rotation calculation means And the difference between the position and rotation of the input fingerprint and the corrected position and rotation, the similarity is calculated by the similarity calculation means, and the registered fingerprint owner and the input are input according to the calculated similarity. A determination unit configured to determine whether the owner of the fingerprint is the same person; and an output unit configured to output a determination result determined by the determination unit.

  According to the computer system of the present invention, when obtaining the difference between the movement amount and the rotation amount of the registered fingerprint and the input fingerprint, first, the approximate values thereof are obtained, and then the detailed difference between the movement amount and the difference between the rotation amounts. And ask. By using such a multistage collation process, the fingerprint collation process is increased as compared with the conventional fingerprint collation process. However, since the work time in each process can be reduced, the verification time can be shortened as a whole. In addition, according to the present invention, the amount of work in each process can be reduced as compared with the conventional fingerprint collation method, so that the memory of the computer system can be reduced. Specifically, as shown in Table 3 of the present specification, according to the small memory matching algorithm of the present invention, fingerprint matching is performed with a memory of 1/5/5 compared to the conventional basic algorithm. You can see that it can be done.

  In this specification, “the approximate value of the difference between the registered fingerprint position and the input fingerprint position” means that the computer system for fingerprint collation according to the present invention determines the final registered fingerprint position and the input fingerprint position. This is a value obtained before determining the difference value, and is a difference value that is less accurate than the final difference value. Such an approximate value is a value for attaching a “memboshi” for determining an approximate value of the final difference value. For example, the approximate value has a range of about 5 to 200 times the final difference value. An area corresponding to the difference value within the specified range may be obtained and used as an approximate value.

  In this specification, “the approximate value of the difference between the rotation of the registered fingerprint and the rotation of the input fingerprint” means that the computer system for fingerprint verification according to the present invention performs the final rotation of the registered fingerprint and the rotation of the input fingerprint. This is a value obtained before determining the difference value, and is a difference value that is less accurate than the final difference value. Such an approximate value is a value for attaching a “memboshi” for determining an approximate value of the final difference value. For example, the approximate value has a range of about 5 to 200 times the final difference value. An area corresponding to the difference value within the specified range may be obtained and used as an approximate value.

  In the present specification, the “detailed value of the difference in position between the registered fingerprint and the input fingerprint and the difference in rotation” means the final value of the difference value in the computer system for fingerprint verification of the present invention. This is a value for determining the identity between the registered fingerprint holder and the input fingerprint holder. Therefore, the “detailed value of the difference in the position of the registered fingerprint and the input fingerprint and the difference in the rotation” only needs to be accurate enough to perform fingerprint verification in a computer system for fingerprint verification. Any accuracy may be used as long as the accuracy is as required by publicly known literature, or an accuracy exceeding that accuracy. In the present invention, the “detailed value of the difference in position between the registered fingerprint and the input fingerprint and the difference in rotation” is preferably obtained by further dividing the vicinity of the approximate value obtained previously.

  [2] Another aspect of the computer system for fingerprint collation according to the present invention is a registration stored in the fingerprint information storage means using the similarity calculation means instead of the first position / rotation calculation means. Examples include a first position calculation unit that obtains an approximate value of a difference between a fingerprint position and an input fingerprint position. In this embodiment, “a detailed value of a difference in position between a registered fingerprint and an input fingerprint and a difference in rotation” is obtained without obtaining an approximate value of rotation. Further, in this embodiment, the detailed value of the difference in rotation is not based on the approximate value, and the detailed value is obtained by the second rotating means. Since the computer system of this embodiment first obtains an approximate value of the difference in position between the registered fingerprint and the input fingerprint, only a small working memory is required, and the processing speed becomes high.

  [3] The feature point of the fingerprint in the computer system of the present invention may be one or both of an end point and a branch point, and the information related to the feature point of the fingerprint is the position information of the end point of the feature point, and The information about the angle formed by the lines whose main points are the sampling points provided so as to have a predetermined interval on the feature point is given.

  [4] In order to solve the above-described problem, the fingerprint collation system of the present invention includes a fingerprint collation system including the computer system of the present invention.

  Since this fingerprint collation system has the above-described computer system, a small working memory is required and the processing speed is high.

  [5] More specifically, the fingerprint collation system of the present invention binarizes a filter unit that removes noise from fingerprint information read by a fingerprint sensor and digitized, and fingerprint information from which noise has been removed by the filter unit. A pre-processing unit including a binarization circuit that performs processing, a thinning unit that thins the fingerprint information binarized by the binarization circuit, and fingerprint image information processed by the pre-processing unit. There is a fingerprint collation system that includes a feature extraction unit that extracts information on a database, a database unit that stores information on feature points of a fingerprint extracted by the feature extraction unit, and the computer system of the present invention.

  In the fingerprint collation system of this embodiment, since predetermined preprocessing can be performed, fingerprint collation can be performed more accurately.

  [6] The mobile phone of the present invention is a mobile phone including the fingerprint verification system described above. Since this mobile phone includes a fingerprint verification system, it is possible to prevent a situation in which information stored in the mobile phone is stolen by another person. In mobile phones, there is a strong demand for miniaturization. Therefore, since the fingerprint collation system of the present invention can make the memory smaller than the conventional one, the mobile phone of the present invention can effectively meet such a request.

  When the computer system of the present invention obtains the difference between the registered fingerprint and the difference between the amount of movement and the amount of rotation of the input fingerprint, it first obtains an approximate value thereof, and then the detailed difference between the amount of movement and the difference between the amounts of rotation. Since it is based on an effective and unprecedented algorithm for obtaining the above, it is possible to achieve high-precision fingerprint collation with less memory than in the past.

  In addition, the computer system of the present invention first obtains approximate values when obtaining the difference between the registered fingerprint and the difference between the movement amount and the rotation amount of the input fingerprint, and then obtains the detailed difference between the movement amount and the rotation amount. Since a circuit based on an effective algorithm for obtaining the difference is provided, fingerprint matching with higher accuracy can be achieved more quickly than in the past.

  Since the fingerprint collation system of the present invention includes the computer system as described above, fingerprints can be collated quickly and accurately, and the working memory is small.

  Since the fingerprint collation system of the present invention performs predetermined pre-processing and includes the above-described computer system, it provides an advisory collation system that takes into account differences in fingerprint images in the fingerprint registration process and the fingerprint collation process. be able to.

(1. Computer system for fingerprint verification)
As shown in FIG. 3, the computer system for fingerprint collation according to the present invention (hereinafter also simply referred to as “computer system”) includes, for example, an input / output port 12 that is electrically connected to the outside, and a read-only port. A memory 13, a work memory 14, a collation unit 15 such as a processor for controlling work such as collation, and a bypass 16 that electrically connects them. In addition, what comprises the memory | storage part 17 with the read-only memory 13 and the working memory 14 is a preferable embodiment of this invention.

(1.1. I / O port)
The input / output port 12 is electrically connected to, for example, a database in which feature information is recorded, a feature extraction unit 4 of the fingerprint collation system 1, an external output unit such as a monitor, a computer, etc. A port for input or output. The input / output port 12 functions as input means and output means of the present invention. The input means and output means of the present invention may be the same as the input / output port, or may be provided separately.

(1.2. Read specialized memory)
The read specialized memory is a memory that stores feature information input from the input / output port 12 in a readable manner. The read specialized memory is composed of, for example, a RAM.

(1.3. Working memory)
The work memory 14 is a memory used for work such as collation. The work memory 14 includes, for example, a ROM and a RAM. And the memory | storage part 17 which consists of read-only memory 13 and the working memory 14 functions as a memory | storage means of this invention.

(1.4. Verification unit)
The collation unit 15 compares the registered fingerprint feature information with the input fingerprint feature information according to a predetermined algorithm, and determines whether the registered fingerprint owner and the input fingerprint owner are the same person. It is a part which determines whether. In addition, the collation part 15 functions as a collation means of this invention.

  FIG. 4 is a block diagram for explaining an example of the matching unit 15 of the present invention. As shown in FIG. 4, the collation unit 15 of the present invention includes, for example, a similarity calculation unit 18, a first position / rotation calculation unit 19, a second position / rotation calculation unit 20, and a determination unit 21. It comprises. Since the computer system of the present invention has such a configuration as the registered fingerprint, when calculating the difference between the movement amount of the input fingerprint and the difference between the rotation amounts, first, the approximate values are obtained, and then the detailed movement amount. And the difference in rotation amount. For this reason, the computer system of the present invention can perform high-precision fingerprint collation at a high speed with less memory than in the past.

(1.4.1. Similarity calculation unit)
The similarity calculation unit 18 is a unit that obtains the similarity between two fingerprints from information (feature information) on the input feature points of the two fingerprints, and functions as a similarity calculation unit of the present invention.

(1.4.2. First position / rotation calculation unit)
The first position / rotation calculation unit 19 uses the similarity calculation unit 18 to obtain a rough value of the difference between the registered fingerprint position and the input fingerprint position stored in the fingerprint information storage unit. And the difference between the position of the registered fingerprint calculated by the first position calculation unit and the position of the input fingerprint, and then using the similarity calculation unit 18, the rotation of the registered fingerprint and the rotation of the input fingerprint are performed. A first rotation calculation unit for obtaining an approximate value of the difference, and a unit for obtaining an approximate value of a difference in position between the registered fingerprint and the input fingerprint and a difference in rotation, the first position / rotation calculation of the present invention Functions as a means.

Note that, instead of the first position / rotation calculation means, the similarity calculation means is used to obtain an approximate value of the difference between the registered fingerprint position and the input fingerprint position stored in the fingerprint information storage means. It is another preferred embodiment of the present invention that includes first position calculating means for obtaining an approximate value of the difference in position between the registered fingerprint and the input fingerprint. Such a computer system for fingerprint verification first calculates the approximate value of the difference in finger position between the fingerprint registration process and the verification process, and then calculates the detailed difference between the position and rotation, thus reducing the working memory. Can also shorten the verification time.

(1.2.3. Second position / rotation calculation unit)
The second position / rotation calculation unit 20 uses the approximate value of the position difference and rotation difference between the registered fingerprint and the input fingerprint calculated by the first position / rotation calculation unit 19 and the similarity calculation unit 18. The second position calculation for obtaining the difference between the registered fingerprint position stored in the fingerprint information storage unit and the input fingerprint position in a narrower range than the first position calculation unit in the vicinity of the approximate value. , The information about the difference between the position of the registered fingerprint calculated by the second position calculation unit and the position of the input fingerprint, and the similarity calculation unit 18, and the rotation of the registered fingerprint stored in the fingerprint information storage unit And a second rotation calculation unit for determining in detail the difference between the rotation of the input fingerprint and the rotation of the input fingerprint in a range narrower than the first rotation calculation unit in the vicinity of the approximate value, and the position of the registered fingerprint and the input fingerprint Detailed value of the difference in rotation and the difference in rotation A Mel unit, functions as a second position and rotation calculating means of the present invention.

(1.4.4. Determination unit)
The determination unit 21 corrects the difference in position and rotation between the registered fingerprint calculated by the second position / rotation calculation unit 20 and the input fingerprint, and makes the similarity calculation unit similar in the corrected position and rotation. It is a unit that calculates the degree and determines whether the owner of the registered fingerprint and the owner of the input fingerprint are the same person according to the calculated similarity, and functions as a determination unit of the present invention.

(2. Fingerprint verification system)
FIG. 5 shows an example of the fingerprint collation system 1 of the present invention. As shown in FIG. 5, the fingerprint collation system 1 includes, for example, a fingerprint sensor 2, a preprocessing unit 3, a feature extraction unit 4, a database unit 5 that stores a plurality of templates 6, a computer system 11, And an external display unit 8. Since the fingerprint collation system of the present invention uses the computer system described above, fingerprints can be collated quickly and accurately, and the working memory is small. A fingerprint collation system that can be connected to the fingerprint sensor 2 and does not include the fingerprint sensor 2 is another preferred embodiment of the present invention.

(2.1. Fingerprint sensor)
The fingerprint sensor 2 is a means for reading a fingerprint or a means for performing a predetermined process such as a digitization process after reading the fingerprint. That is, the fingerprint sensor 2 includes an A / D conversion circuit, and it is a preferred embodiment of the present invention to convert analog information relating to a fingerprint image read by the fingerprint sensor 2 into digital information. As the fingerprint sensor, a known sensor used in a fingerprint verification system such as an optical sensor, a capacitance sensor, a pressure sensor, a heat sensor, an electric field sensor, or an ultrasonic sensor can be used.

(2.2. Pre-processing unit)
The preprocessing unit 3 is a unit that performs predetermined preprocessing before the feature extraction unit 4 extracts feature points, and functions as a preprocessing unit. Examples of the preprocessing unit include a filter unit, a binarization circuit, a direction filter unit, and a thinning unit. FIG. 6 is a photograph in place of a drawing showing fingerprints before and after pre-processing. 6A is a gray level fingerprint image, FIG. 6B is a binary image fingerprint image binarized into black and white, and FIG. 6C is a thinned fingerprint image. Represents an image.

(2.2.1. Filter part)
The filter unit is a unit that removes noise by performing predetermined filter processing, and functions as a filter unit. As a filter part, what contains a high pass filter, a low pass filter, etc. is mentioned, for example. For example, a fingerprint image read by a fingerprint sensor and converted into digital information by an A / D conversion circuit contains a lot of noise. For this reason, the fingerprint image becomes a smooth image by being pre-filtered by the high-pass filter and the low-pass filter. FIG. 6A is a gray level image of a fingerprint after such preliminary filtration.

(2.2.2. Binary circuit)
The binarization circuit performs a process of converting the gray level image into a binary value of “0” and “1” with a predetermined threshold as a reference. As the binarization circuit, a circuit capable of binarizing the fingerprint image by selecting a preferable threshold from a plurality of thresholds for the gray level fingerprint image is preferable.

  The binarization circuit is preferably macroscopically dense and microscopically converts fingerprint information on a plurality of pixels regularly arranged at a predetermined interval into white (“1”). A process of binarizing to black (“0”) is performed. As such a binarization circuit, for example, a pixel value of a fingerprint image at a grid point of a 256-gradation grid is used as binarization information of “1” and “0” by a threshold value.

(2.2.3 Direction filter part)
The direction filter unit is, for example, a filter unit that reduces the noise of the gray level image by calculating the maximum direction of the fingerprint from each binarized image block and masking the direction filter along the calculated maximum direction. Yes, it functions as direction filter means. Note that the above-described filter unit may also serve as the direction filter unit.

(2.2.4. Thinning part)
The thinning section is a section for thinning the fingerprint image binarized by the binarization circuit, and functions as the thinning means of the present invention.

(2.2.5. Feature Extraction Unit)
The feature extraction unit 4 performs processing for extracting feature information from fingerprint information and functions as a feature extraction unit of the present invention. The feature points extracted by the feature extraction unit 4 include information on the branch point 9 as shown in FIG. 2A and the end point 10 shown in FIG. Examples of the information on the feature points extracted by the feature extraction unit include information on the position of the start point of the feature points, a vector representing the sampling points 23 on the feature points (an angle formed by the vectors), and the like.

(2.3. Database section)
The database unit 5 is a unit that stores several to dozens of pieces of feature information extracted by the feature extraction unit 4 as a template 6 and functions as a template 6 storage unit of the present invention. It is a preferred embodiment of the present invention that a known relational database is constructed in the database unit 5 and feature information is stored in a hierarchical manner. The database unit 5 may be a memory such as a known ROM or RAM, or may be implemented by software such as a database.

(2.4. Computer system)
As the computer system 11, the computer system of the present invention described above can be used. Since the fingerprint collation system 1 of the present invention uses the computer system of the present invention, fingerprints can be collated quickly and accurately, and the work memory can be reduced.

(2.5. Output display section)
The output display unit is a unit that displays a result of the personal identification determined by the computer system 11, and functions as an output display unit of the present invention. Examples of the output display unit include a known display.

(3. Fingerprint verification method)
FIG. 7 is a chart showing an example of a fingerprint collation method using the fingerprint collation system of the present invention.
The fingerprint collation method of the present invention includes a gray level image acquisition step, a preliminary filtration step, a first binarization step, a direction detection step, a gray level image direction filter step, and a second binarization. A process, a thinning process, a feature point extraction process, and a feature point matching process.

(3.1. Gray level image acquisition process)
The gray level image acquisition step is a step of acquiring a gray level image of a fingerprint by reading the fingerprint with a fingerprint sensor.

(3.2 Preliminary filtration step)
In the preliminary filtration step, noise is removed from the gray level fingerprint image read by the fingerprint sensor and converted into digital information by the A / D conversion circuit. The fingerprint image read by the fingerprint sensor and converted into digital information by the A / D conversion circuit contains a lot of noise. For this reason, the fingerprint image becomes a smooth image by being pre-filtered by the high-pass filter and the low-pass filter. An example of the gray level image obtained in this way is shown in FIG. Examples of the filter used in the preliminary filtration step include a high-pass filter and a low-pass filter. This preliminary filtration step is an optional step, and omitting the preliminary filtration step is another embodiment of the fingerprint matching method of the present invention. Further, it is another embodiment of the fingerprint collation method of the present invention to directly perform a gray level image direction filtering step described later after the preliminary filtering step.

(3.3. First binarization step)
In the first binarization step, the binarization circuit uses the gray level image from which noise has been removed in the preliminary filtration step as black (“0”) and white (“1”) with a predetermined threshold as a reference. ) To convert to binary. In addition, it is another embodiment of the fingerprint collation method of the present invention to extract a feature point by directly thinning a binarized image binarized in this way.

(3.4. Direction detection process)
In the direction detection step, the maximum direction of the ridge is calculated from the block of the binarized image.

(3.5. Gray-level image direction filtering step)
The gray level image direction filtering step is a step of reducing noise of the gray level image by masking the direction filter in accordance with the maximum direction detected by the above direction detection.

(3.6. Second binarization step)
In the second binarization step, the binarization circuit uses, for example, a gray level image whose noise has been reduced through a gray level image direction filtering step or the like as black ("0") with a predetermined threshold as a reference. And white (“1”) are converted into binary values. An example of a binarized image obtained through such a process is shown in FIG. As a method for binarizing a gray level image, “Fingerprint enhancement by directional Fourier filtration” G. Sherlock, D.M. M.M. Monro and K.M. Millard; IEEE Proceeding-Vision, Image and Signal Proceeding, 141 (20): 87-94, 1994, and “enhancement and characteristic refinement of fingerprint images” C. Douglas Hung; IEEE Trans. See On Pattern Recognition, 26 (11): 1661-1671, 1994. Emphasizing a fingerprint image by binarization reduces noise and sharpens ridges.

(3.7. Thinning process)
In the thinning process, the thinning circuit performs a process of thinning the binarized image obtained in the second binarization process. An example of a thinned fingerprint image obtained in this way is shown in FIG. Examples of the thinning method include a method of tracing the center of a binarized ridge. By performing the thinning process in this way, the skeleton of the ridge pattern of the fingerprint becomes clear.

(3.8. Feature point extraction process)
The feature point extraction step is, for example, a step of extracting information about feature points (feature information) from a thinned binary image. An example of feature information extracted in this feature point extraction step will be described below. FIG. 8 is a diagram for explaining a process for extracting feature information from feature points. FIG. 8A is a diagram showing the relationship between the end points of feature points and sample points on the feature points. FIG. 8B is a diagram showing the relationship between the end points of the feature points and the parameters extracted from the sampling points 23.

(3.8.1. Feature point search process)
First, the feature extraction unit 4 searches the fingerprint image for a branch point 9 as shown in FIG. 2A and an end point as shown in FIG. 2B (a feature point search step).

(3.8.2. Sampling point calculation step)
In the sampling point calculation step, the feature extraction unit 4 calculates a point on the feature point that is a distance D away from the end point 22 of the found feature point as the first sampling point 23, and the first sampling point 23, a point on the feature point that is a distance D away is set as the second sampling point 23, and in the same manner, four sampling points 23 on the feature point are calculated (sampling point calculation step). An example of the sampling points 23 calculated on the feature points in this way is shown in FIG. The distance D may be determined based on the balance between the fingerprint size and the required accuracy. In the above example, the number of sampling points is four. However, the number is not limited to four, and may be three to six, may be five, or may be more than that. There may be. However, if the number is too large, the accuracy increases slightly, but the working memory required for the processing also increases, and the processing time increases.

(3.8.3. Parameter calculation step)
In the parameter calculation step, as shown in FIG. 8B, the feature extraction unit 4 performs the ridge with respect to the horizontal axis at the position (Px, Py) of the end point 22 of the feature point and the end point 22 of the feature point. The angle θ of the line direction and the angles (Δφ ₁ , Δφ ₂ , Δφ ₃ ) of two line segments that end at the sampling point 23 calculated in the sampling point calculation step are calculated. The information (MS) related to the feature points shown in FIG. 8B can be parameterized as, for example, the following equation.

Characteristic information (MS): s = (Px, Py, θ, Δφ ₁ , Δφ ₂ , Δφ ₃ )

In the registration process, such parameterization is performed for a plurality of feature points, and each parameter is stored in the database unit 5 as a template 6. Hereinafter, in this specification, information (feature information) related to the feature points registered as the template 6 is represented as “t _i ”, and a set of templates for a certain individual is represented as “T _m ” (1 ≦ i ≦ m). . That is, if the number of templates for an individual is m, “T _m ” can be expressed as follows using “t _i ”.

T _m = {(t ₁ , t ₂ ,..., T _m ) | t _i ∈MS}

On the other hand, the feature points of the input fingerprint input for collation are also parameterized. The parameterized input fingerprint feature information is represented as “s _j ”, and the set of input fingerprint feature information is represented as “I _n ” (1 ≦ j ≦ n). That is, if the number of feature information of the input fingerprint is n, “I _n ” can be expressed as follows using “s _j ”.

I _n = {(s ₁ , s ₂ ,..., S _n ) | s _i ∈ MS}

(3.9. Feature point matching process)
The feature point matching process uses the registered fingerprint feature information and the input fingerprint feature information to correct differences in the position and rotation between the registered fingerprint and the input fingerprint, and calculates the corrected registered fingerprint and input fingerprint similarity. In this step, it is determined whether or not the registered fingerprint owner and the input fingerprint owner are the same person.

(3.9.1. Similarity calculation method)
The similarity is calculated by, for example, calculating the similarity between the angle information (for example, Δφ ₁ , Δφ ₂ , Δφ ₃ ) of the corresponding feature points included in the above “T _m ” and “I _n ”. Can be sought. A value obtained by calculating a pair of similarities between t _i and s _j of the feature information (t _i , s _j εMS) at each feature point is referred to as Sim (t _i , s _j ).

(3.9.2. Position approximate value calculation step)
The first position calculating unit, for each of the feature points included in the I _n, obtains the degree of similarity between all the feature points included in the T _m, accumulating the deviation of the position of the pair of similarity is maximized. FIG. 9 is a diagram illustrating a shift in the position of the feature points included in “T _m ” where the degree of similarity is maximum and “I _n ”. The vertical axis represents the displacement in the y-axis direction, and the horizontal axis represents the displacement in the x-axis direction. FIG. 9 also shows the bar graphs (histograms) (HistoX, HistoY) in which the deviations in the x-axis direction and the y-axis direction are accumulated for each predetermined range. For the predetermined range, a quantized value is used instead of every pixel. Then, the first position calculation unit obtains a value that gives the maximum value of the deviation in the X-axis direction and the deviation in the Y-axis direction as an approximate value of the deviation between the registered fingerprint and the input fingerprint (position approximate value calculation step). ). The first position calculation unit functions as a first position calculation unit of the present invention.

  FIG. 10 is a diagram illustrating how the parameterized feature points are moved. FIG. 10A shows a state where the end point (Px, Py) of the feature point is translated by (ΔPx, ΔPy). The point (Px ′, Py ′) obtained by translating the end point (Px, Py) of the feature point by (ΔPx, ΔPy) can be calculated as follows.

Px ′ = Px + ΔPx
Py ′ = Py + ΔPy

(3.9.3. Calculation of approximate rotation value)
After the position approximate value calculation step, the difference between the position and rotation of the registered fingerprint and the input fingerprint is calculated in detail. After the position approximate value calculation step, the first rotation calculation unit calculates the first position calculation unit. After correcting for the difference between the position of the registered fingerprint and the position of the input fingerprint, and using the similarity calculation unit 18 to roughly determine the difference between the rotation of the registered fingerprint and the rotation of the input fingerprint, Embodiment.

  FIG. 10B shows a state in which the feature point (Px, Py, θ) is rotationally moved by (Δθ). The points (Px ′, Py ′) obtained by rotating the feature points (Px, Py, θ) by (Δθ) can be calculated as follows.

Px ′ = Pxcos (Δθ) −Pysin (Δθ)
Py ′ = Pxsin (Δθ) + Pycos (Δθ)

Even if the feature points are translated or rotated, (Δφ ₁ , Δφ ₂ , Δφ ₃ ) of the feature information remains unchanged. Therefore, the degree of similarity can be obtained from the angle information. This approximate rotation value calculation step is an optional step and may be omitted.

(3.9.4. Position calculation step)
In the position calculating step, first, the second position calculating unit performs the difference in position between the registered fingerprint calculated by the first position / rotation calculating unit 19 or the first position calculating unit and the input fingerprint (and rotation rotation). The difference between the registered fingerprint position and the input fingerprint position stored in the fingerprint information storage unit is corrected using the corrected registered fingerprint feature information and the input fingerprint feature information. It is obtained in detail in a narrow range compared to the first position calculation unit in the vicinity of the value. Note that the second position calculation unit functions as a second position calculation unit of the present invention. In this position calculation step, for example, in the area of 10 × 10 pixels, for example, with the approximate value of the position difference obtained previously as the center,
Sim (t _i , s _j ) is obtained for each pixel while shifting the position. The “neighboring approximate value” required in this position calculation process is not limited to the conventional fingerprint verification method, as long as it does not cover all areas of the fingerprint image. However, for example, 9 pixels or more and 10,000 pixels or less are mentioned, preferably 16 pixels or more and 2500 pixels or less, more preferably 36 pixels or more and 1600 pixels or less, and further preferably 49 pixels or more and 900 pixels or less. Yes, particularly preferably 81 pixels or more and 121 pixels or less.

  In this position calculation step, for example, a set of data including the values of ΔPx and ΔPy and the similarity is obtained, and among them, (ΔPx, ΔPy) giving the highest similarity is finally registered. It is calculated as the difference in fingerprint position between the process and the verification process.

(3.9.5. Rotation calculation step)
The rotation calculation step is a step of calculating a difference in fingerprint rotation between the registration step and the collation step. In the rotation calculation process, for example, after correcting the difference in the position of the fingerprint in the registration process obtained in the position calculation process and the collation process, correction for adding any rotation movement of the registered fingerprint and input fingerprint data is performed. Apply. As a result, for example, a set of (Δθ) and the corresponding similarity is obtained, and (Δθ) giving the maximum similarity among (Δθ) is the rotation of the fingerprint in the registration step and the matching step. Calculated as the difference. It should be noted that when the approximate value of the difference in rotation has been obtained in the previous rotation approximate value calculation step, the similarity may be calculated by shifting by a small angle.

(3.9.6. Similarity calculation step)
The similarity calculation step is a step of calculating the fingerprint similarity in order to determine the identity of the registered fingerprint owner and the verification fingerprint owner. The similarity calculation step includes, for example, a difference in fingerprint position between the registration step obtained in the position calculation step and the collation step, a registration step obtained in the rotation calculation step, and a fingerprint in the collation step. And the similarity between the corrected registered fingerprint and the verification fingerprint is calculated.

(3.9.7. Determination step)
The determination step is a step of determining the identity of the registered fingerprint owner and the verification fingerprint owner based on the similarity between the registered fingerprint calculated in the similarity calculation step and the verification fingerprint. For example, the determination process determines that the registered fingerprint owner and the input fingerprint owner are the same if the similarity exceeds a predetermined threshold, and if the similarity does not exceed the predetermined threshold, One that determines that the owners of the input fingerprints are not identical.

(4. Performance comparison)
In fingerprint recognition, when comparing a registered fingerprint and an input fingerprint, a rough value is not obtained for the difference in position between them, and (ΔPx, ΔPy) is shifted for all pixels, and a similarity is obtained. An algorithm that determines the difference in rotation between the registered fingerprint and the input fingerprint is used as a basic algorithm. This collation algorithm can be expressed as shown in Table 1 below, for example.

  In fingerprint recognition, when comparing a registered fingerprint and an input fingerprint, an approximate value is obtained for the difference in position between them, and thereafter, a similarity obtained by shifting (ΔPx, ΔPy) for pixels in a specific range is obtained. An algorithm that obtains the maximum similarity is defined as a position shift, and an algorithm for obtaining a difference in rotation between the registered fingerprint and the input fingerprint is referred to as a small memory matching algorithm. This collation algorithm can be expressed as shown in Table 2 below, for example.

  In the following, the performance comparison between the basic algorithm and the small memory matching algorithm will be described. In the following, the size of the fingerprint image is set to 300 × 300 pixels. In the small memory matching algorithm, ΔPx and ΔPy are each quantized to 8 pixels. In the small memory matching algorithm, Δθ is quantized in 32 directions. The number of feature points of the input fingerprint and the number of registered fingerprint templates was 25. Each was divided into 11 pieces in the X-axis and Y-axis directions of the small memory verification algorithm. The frequency of the fingerprint verification system is 100 MHz, the processing time of the rotation Sj is 200 clock cycles, the processing time for updating the memory calculation to another is 140 clock cycles, and the processing time for checking one item when searching for the peak value is 15 Executed in the clock cycle. Table 3 below shows a comparison of the performance of the basic algorithm and the small memory matching algorithm based on the above assumption.

  From Table 3, it can be seen that the small memory verification algorithm, which is an algorithm employed in the computer system of the present invention, requires less work memory and can shorten the work time.

  The computer system for fingerprint collation according to the present invention requires less memory and can collate fingerprints at high speed and with high accuracy, so access to IC chips with fingerprint collation, cards, computer networks, bank cards, mobile phones, and homes. It can be used for various devices such as devices that use fingerprint verification.

FIG. 1 is a diagram illustrating an example of a fingerprint collation system. FIG. 2 is a diagram illustrating an example of feature points of a fingerprint. FIG. 2A is a diagram showing a part of a fingerprint having a branch point and FIG. 2B a feature point being an end point. FIG. 2C is a photograph replacing a drawing showing a fingerprint including feature points. FIG. 3 is a block diagram showing an embodiment of the fingerprint matching system of the present invention. FIG. 4 is a block diagram for explaining an example of the collation unit of the present invention. FIG. 5 shows an example of the fingerprint collation system of the present invention. FIG. 6 is a photograph in place of a drawing showing fingerprints before and after pre-processing. 6A is a gray level fingerprint image, FIG. 6B is a binary image fingerprint image binarized into black and white, and FIG. 6C is a thinned fingerprint image. Represents an image. FIG. 7 is a chart showing an example of a fingerprint collation method using the fingerprint collation system of the present invention. FIG. 8 is a diagram for explaining a process for extracting feature information from feature points. FIG. 8A is a diagram showing the relationship between the end points of feature points and sample points on the feature points. FIG. 8B is a diagram illustrating the relationship between the end points of the feature points and the parameters extracted from the sampling points. FIG. 9 is a diagram illustrating a shift in the position of the feature points included in “T _m ” where the degree of similarity is maximum and “I _n ”. FIG. 10 is a diagram illustrating how the parameterized feature points are moved. FIG. 10 (a) represents a parallel movement (correcting the position difference), and FIG. 10 (b) represents a rotational movement.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Fingerprint collation system 2 Fingerprint sensor 3 Preprocessing part 4 Feature extraction part 5 Database part 6 Template 7 Collation part 8 External display part 9 Fingerprint branch point 10 Fingerprint end point 11 Computer system 12 Input / output port 13 Read-only memory 14 Work memory DESCRIPTION OF SYMBOLS 15 Collation part 16 Bypass 17 Memory | storage part 18 Similarity calculation part 19 1st position / rotation calculation part 20 2nd position / rotation calculation part 21 Judgment part 22 End point of feature point 23 Sampling point

Claims (6)

An input means for inputting information about the feature points of the fingerprint;
Fingerprint information storage means for storing information on feature points of the registered fingerprint input by the input means and information on feature points of the input fingerprint input by the input means;
A similarity calculation means for obtaining a similarity between these two fingerprints from information related to the feature points of the registered fingerprint and information related to the feature points of the input fingerprint;
First position calculating means for obtaining an approximate value of a difference between a registered fingerprint position stored in the fingerprint information storing means and an input fingerprint position using the similarity calculating means, and the first position calculating means A first rotation calculating means for correcting a difference between the position of the registered fingerprint and the position of the input fingerprint, and obtaining an approximate value of the difference between the rotation of the registered fingerprint and the rotation of the input fingerprint using the similarity calculating means; A first position / rotation calculating means for obtaining an approximate value of a difference in position between the registered fingerprint and the input fingerprint and a difference in rotation;
Using the approximate value of the position difference and the rotation difference between the registered fingerprint and the input fingerprint calculated by the first position / rotation calculation means and the similarity calculation means, the registered fingerprint stored in the fingerprint information storage means The second position calculation means, which calculates the difference between the position and the position of the input fingerprint in a narrower range than the first position calculation means in the vicinity of the approximate value, and the second position calculation means Using the information on the difference between the position of the registered fingerprint and the position of the input fingerprint and the similarity calculation means, the difference between the rotation of the registered fingerprint and the rotation of the input fingerprint stored in the fingerprint information storage means is in the vicinity of the approximate value. A second rotation calculation means for obtaining a detailed value in a narrower range than the first rotation calculation means, and for obtaining a detailed value of a difference in position between the registered fingerprint and the input fingerprint and a difference in rotation. Second position / rotation And it means out,
The position and rotation difference between the registered fingerprint and the input fingerprint calculated by the second position / rotation calculation means is corrected, and the similarity is calculated by the similarity calculation means at the corrected position and rotation. A determination means for determining whether the owner of the registered fingerprint and the owner of the input fingerprint are the same person according to the similarity degree,
Output means for outputting the determination result determined by the determination means;
A computer system for fingerprint verification that performs fingerprint verification.   Instead of the first position / rotation calculation means, a first position for obtaining an approximate value of the difference between the registered fingerprint position and the input fingerprint position stored in the fingerprint information storage means using the similarity calculation means The computer system according to claim 1, further comprising a calculation unit. The feature points of the fingerprint are either or both of end points and branch points,
The information related to the feature points of the fingerprint extracted by the feature extraction means includes the position information of the end points of the feature points, and a line whose main point is a sampling point provided at a predetermined interval on the feature points. The computer system according to claim 1, wherein the computer system is information relating to an angle formed by each other.   The fingerprint collation system which comprises the computer system in any one of Claims 1-3. Fingerprint sensor read and digitized filter information to remove noise, binarization circuit to binarize fingerprint information from which noise has been removed by filter section, and binarization by binarization circuit A preprocessing unit including a thinning unit for thinning fingerprint information;
A feature extraction unit that extracts information about feature points of the fingerprint from the fingerprint image information processed by the preprocessing unit;
A database unit for storing information on feature points of the fingerprint extracted by the feature extraction unit;
A fingerprint verification system comprising: the computer system according to any one of claims 1 to 3. A mobile phone including the fingerprint matching system according to claim 4 or 5.