JPH06103366A - Fingerprint collating method/device - Google Patents

Abstract

PURPOSE:To quickly perform the collation of fingerprints at a high recognition rate by collating the fingerprint image data with the registered data via a template. CONSTITUTION:The fingerprint collating device is consisted of a fingerprint image input device 11, an A/D converter 12, a frame memory 13 which stores the image data, an image operational processor 20, a ROM 21 which stores the adjacent point pattern guidance value used for the thin-lined scan, a ROM 22 to be used for a feature point extraction scan a memory 23 which stores the control information on the thin-lined scan, a memory 41 which stores the registered data, an input device 42, an output device 43, an access signal generator 44, and a CPU 40. In this case, a digital signal processor fitted to the image processing arithmetic operation is provided independently of the CPU 40, then the thin-lined processing and the feature point extracting processing can be carried out at a high speed in the unit of eight nearby cells. Then, the fingerprints lacking for outstanding feature point can be collated at a high recognition rate by a template collation method.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fingerprint collation device and a fingerprint collation method, and more specifically, it collates and compares collated fingerprint image data with registered fingerprint image data in real time, and selects access or access based on the determination result. The present invention relates to a fingerprint collation device and a fingerprint collation method for controlling.

[0002]

2. Description of the Related Art Fingerprints are "lifelong" and "universal"
It is used as the most powerful means of personal authentication due to these two major characteristics. The feature points used for fingerprint matching are
A ridge (an end point) of a ridge (a convex portion of a fingerprint pattern) and a point where the ridge branches (branch point) are collectively referred to as a feature point (minuia). Matching a fingerprint as an image poses a problem in terms of file size and matching speed, so it is realized by matching the positional relationship of feature points extracted from the fingerprint image. In the feature extraction process of the input fingerprint image, the fingerprint source image is subjected to image correction including smoothing and enhancement, and then the directionality of the pattern is extracted over the entire image. A binarized ridge image is extracted by an anisotropic local two-dimensional differential filter that depends on the extracted directionality and threshold processing. The ridge line image is converted into a cord and the position of the feature point is detected by the 3 × 3 detection mask, and the direction of the ridge line is measured for each feature point. Furthermore, by detecting the number of intersecting ridges as the relationship between neighboring feature points, the positional coordinates of each feature point, the pattern direction, and the relationship between neighboring feature points are listed as fingerprint features.
The matching process between the input fingerprint and the registered fingerprint is performed by matching the overall displacement and rotation according to the fingerprint features, and then performing a “pair” inspection for each feature point. As a conventional fingerprint collation device of this type, for example, a fingerprint collation device that performs collation determination of collation fingerprint image data with registered fingerprint image data in real time, and controls selection of access or access based on the determination result. There is. This fingerprint collation device performs preprocessing for removing noise included in an input image and binarization processing for converting a pixel value of image data into a binary value of '1' or '0' at a predetermined threshold. A general-purpose microprocessor performs thinning processing for extracting a line figure with a uniform line width from a binarized image, feature point extraction processing, and image processing operations such as pattern matching.

[0003]

However, in such a conventional fingerprint collation apparatus, since the image processing operation for fingerprint collation is exclusively performed by a general-purpose microprocessor, the amount of data and the complexity of processing are increased. It has been difficult to achieve a desired matching response time because the processing time is lengthened in proportion to. That is, when the image processing is performed by computer software, for example, when the thinning processing is performed on the image data of 512 × 512 pixels in units of 8 × 3 cells of 3 × 3 pixels by sequential scanning, 5
The thinning determination scanning in units of 8 neighboring cells is performed 10 × 510 times. In the thinning process to extract a connected figure of binary image data as a line figure with the same line width without losing its connectivity, the line width should be-and the position of the line should be the center of the original figure. , Maintaining shape connectivity,
Assuming that the end points of the figure are not degenerated unnecessarily, it is necessary to perform processing to suppress the occurrence of whiskers at the irregularities of the boundary and distortion at intersections. The determination process itself also requires a considerable number of steps. The thinning scanning in units of neighboring cells is sequentially and repeatedly repeated until the thinning of the screen converges. Therefore, in order to reduce the response time of the fingerprint collation device, it is necessary to improve the processing method and the hardware configuration for performing the thinning processing in a short time.

If the dedicated hardware for image calculation is mounted, the desired response time is certainly achieved, but the dedicated image hardware is used for the fingerprint collation device as a consumer product which is required to be low in cost and small in size. Is difficult to implement, and the implementation of dedicated image hardware is inferior to the software processing of the computer in terms of expandability.

Further, the conventional fingerprint collation apparatus implements either the feature point collation method or the template collation method in the collation process. This feature point matching method is excellent in that it provides stable matching even for fingerprint rotation deviations, but not everyone has clear fingerprint feature points, and ridge lines are completely collapsed. There are cases where it is not clear. In this case, a higher recognition rate may be obtained with the template matching method than with the feature point matching method, and there is no guarantee that a fingerprint recognition device having only the feature point matching method can always achieve a high recognition rate. An object of the present invention is to provide a fingerprint collation device and a fingerprint collation method for performing collation quickly while maintaining a high recognition rate. Another object of the present invention is to provide a low-cost and small-sized fingerprint matching device and fingerprint matching method.

[0006]

The invention according to claim 1 is
In order to achieve the above-mentioned object, the fingerprint matching image is matched against the registered fingerprint image data and the selection of access or access is suppressed based on the judgment result. The pixel value is binarized into white and black. A thinning processing unit for thinning the binarized image data, the thinning determining unit determining a deletable point by a table search in units of 8 neighboring cells, and scanning for reducing the number of scans of the thinning determination unit. Control means,
Thinning processing means including: and characteristic point extracting means for extracting characteristic points of the image data, and means for extracting characteristic point candidates from the thinned image data by a table search in units of 8 neighboring cells, A characteristic point extracting unit that includes a pseudo characteristic point removing unit that deletes a pseudo characteristic point from the characteristic point candidate;
Matching means for registering true feature points of image data, feature point matching means for determining match between feature points of collated fingerprint image data and registered feature points, template data registration means for template matching, and collated fingerprint image data And template matching means for matching the registered data with the template.

According to a second aspect of the present invention, the thinning determination means is a pre-allocated predetermined position assigned to the position of each adjacent point with respect to the center point in an 8-neighboring cell consisting of the center point and eight adjacent points. Calculating a product of the weighting factor of P and the pixel value of the adjacent point for each of the eight adjacent points, and calculating an adjacent point pattern index value.
For all combination patterns in which the pixel value of the center point of the neighboring cell can be changed from '1' to '0', a table search is performed in the memory device in which the index value of each adjacent point pattern is exhausted, and the adjacent point pattern. Changing the pixel value "1" at the center point of the eight neighboring cells to "0" when the index value exists in the table.

According to a third aspect of the present invention, the scan control means is a first-step screen scan in which a screen of a predetermined size of image data is scanned by the thinning determination means, and all lines of the screen are line by line. To apply the thinning determination means by scanning for each pixel, a flag indicating at least whether the thinning of the scanning line has converged, and a flag of a neighboring cell in which the thinning in the line has not converged. A process of storing scan control data in a memory, which includes position information in a line and the number of neighboring cells in which thinning in the line has not converged continuously, and a screen scanning process in a first step, Second
Screen scanning processing after the step, selected only when the thinning processing of the screen scanning of the step immediately before the step does not converge, based on the scan control data of the step immediately before the step, A screen scanning process in which the total number of lines and pixels to be scanned in the thinning determination process is reduced, a process of updating the contents of the scan control data, and a screen scanning process of the second and subsequent stages,
A convergence determination step of determining whether or not the screen thinning has converged, and a convergence determination step of repeatedly repeating the screen scanning processing from the second step when the screen thinning does not converge. There is.

According to a fourth aspect of the present invention, the feature point extracting means calculates the adjacent point pattern index value of the eight neighboring cells, and the adjacent pattern index value is set to the intersection point, the branch point, and the end point. And a step of searching a table in the memory device in which the point pattern index values are stored, and a step of registering the neighboring cell as a feature point candidate when the same adjacent store pattern value exists in the memory. Are

According to a fifth aspect of the present invention, the pseudo-feature point removing means has the same pair of feature point candidates in the same direction, and the distance between the feature point candidate pairs is equal to or less than a predetermined value. Are deleted from the feature point candidates.

According to a sixth aspect of the present invention, the feature point registering means selects a branch point closest to the center point of the screen data as a parent branch point, and the parent branch point is XY.
In the quadrants of the first quadrant, the second quadrant, the third quadrant, and the fourth quadrant as the origin of the coordinate axes, a step of selecting a branch point closest to the parent branch point as a child branch point of each quadrant, In each quadrant, a step of selecting a predetermined number of branch points as descendant branch points in the order of proximity to child branch points, at least the total number of branch points, branch point position information, parent branch points and child branch points And the mutual positional relationship between the child branch point and the grandchild branch point are registered.

According to a seventh aspect of the present invention, the means for collating the feature points compares the total number of branch points of the collation data with the total number of branch points of the registered data, and determines the branch points in the order of proximity to the center point of the collation data. Selecting a predetermined number of parent branch point candidates, determining whether the collator branch point of the parent branch point is in the vicinity of the corresponding registrar branch point of the registered branch point data, and the collation Matching child branch point: determining whether the grandchild branch point is in the vicinity of the corresponding registered grandchild branch point of the registered branch point data, and the matching child branch point is in the vicinity of the registered child branch point. , The step of scoring the mismatch frequency based on the norm of the collator branch point and the registrant branch point, and the matching grandchild branch point and the registered grandchild branch point when the matching grandchild branch point is in the vicinity of the registered grandchild branch point. The stage of scoring the disagreement frequency based on the norm with the point and the boss Among the average mismatch power stage and the parent branch point candidates for calculating the average mismatch degree about the point candidates,
The method includes a step of selecting a parent branch point of the matching data having a minimum average mismatch frequency, and a step of determining the person as the principal when the average mismatch frequency of the matching parent branch point is a predetermined value or less. .

According to an eighth aspect of the invention, in the process of registering the fingerprint, a step of registering a feature point, a step of collating a feature point between collation image data and registered data, and a template collation data are registered. A step of collating the template of the collation image data and the registration data, and a step of selecting either the feature point collation or the template collation as the collation method of the fingerprint depending on the level of recognition rate of each fingerprint image. I have it.

According to a ninth aspect of the present invention, there is provided a fingerprint collation method for collating collated fingerprint image data with registered fingerprint image data and suppressing entry or access selection based on a determination result. This is a thinning processing step for thinning the binarized image data that has undergone binarization, and the number of scans in the thinning determination step for determining deletable points by a table search in units of 8 neighboring cells and the number of scans in the thinning determination step has been reduced. And a feature point extracting process for extracting feature points of the image data, the feature control process comprising: A feature point extracting step including a step of extracting a point candidate and a pseudo feature point removing step of deleting a pseudo feature point from the feature point candidate, a step of registering a true feature point of an image data, and a collation fingerprint. Image te A feature point matching process, a template data registration process for template matching, and a template matching process for matching the matching fingerprint image data with the registered data are template-matched. .

According to a tenth aspect of the present invention, in the thinning determining step, in an eight-neighbor cell consisting of a center point and eight adjacent points, a predetermined number is assigned to each adjacent point position with respect to the center point. Calculating a product of the weighting factor of the pixel value and the pixel value of the adjacent point for each of the eight adjacent points, and calculating an adjacent point pattern index value,
8 For all combination patterns in which the pixel value of the center point of the neighboring cells can be changed from '1' to '0', a table search is performed in the memory device storing the respective adjacent point pattern index values, and the adjacent point pattern When the index value exists in the table, the pixel value "1" at the center point of the 8 neighboring cells is set to "0".
And the step of changing to.

According to an eleventh aspect of the present invention, the scanning control step is a first-step screen scanning in which a screen of a predetermined size of image data is scanned by the thinning determination step, and all lines of the screen are line by line. To apply the thinning determination means by scanning for each pixel, a flag indicating at least whether the thinning of the scanning line has converged, and a flag of a neighboring cell in which the thinning in the line has not converged. A process of storing scan control data in a memory, which includes position information in a line and the number of neighboring cells in which thinning in the line has not converged continuously, and a screen scanning process in a first step, In the screen scanning process after the second stage, it is selected only when the thinning process of the screen scanning in the stage before the stage does not converge, and is selected based on the scan control data in the stage before the stage. And thinning decision Sense of reduction the screen scanning process the total number of lines and pixels to be scanned and a processing for updating the contents of the scan control data, a second step subsequent screen scanning process consisting,
A convergence determination step of determining whether the screen is thinned or converged, and when the screen thinning does not converge, a convergence determination step of repeatedly repeating the screen scanning processing from the second step onward is provided. There is.

According to a twelfth aspect of the present invention, in the step of extracting the characteristic color, the step of calculating the adjacent-point pattern index value of the eight neighboring cells and the adjacent-pattern index value are set as the adjacency of the intersection, the branch point, and the end point. A step of searching a table in the memory device storing the point pattern index value; and a step of registering the neighboring cell as a feature point catch when the same adjacent store pattern value exists in the memory. ing.

According to a thirteenth aspect of the present invention, in the pseudo feature point removing step, the pairs of feature point candidate pairs of the same type have the same direction, and the distance between the feature point candidate pairs is equal to or less than a predetermined value. Sometimes it is deleted from the feature point candidates.

According to a fourteenth aspect of the present invention, in the step of registering the feature point, a step of selecting a branch point closest to the center point of the screen data as a parent branch point, and the parent branch point are X points.
As the origin of the Y coordinate axis, the first quadrant, the second quadrant, the third quadrant,
In each quadrant of the fourth quadrant, the step of selecting the branch branch point closest to the parent branch point as the child branch short of each quadrant, and in each quadrant, the order of approaching the child branch point Selecting a number of branch points as grandchild branch points, at least the total number of branch points, position information of the branch points,
And a step of registering a mutual positional relationship between the parent branch point and the child branch point and a mutual positional relationship between the child branch point and the grandchild branch point.

According to a fifteenth aspect of the present invention, in the step of collating the characteristic points, the step of comparing the total number of branch points of the collation data with the total number of branch points of the registration data, and the branch points in the order of proximity to the center point of the collation data. Selecting a predetermined number as parent branch point candidates, determining whether the matching branch point of the parent branch point is in the vicinity of the corresponding registrar branch point of the registered branch point data, The step of determining whether the matching grandchild branch point of the matching branch point is within the vicinity of the corresponding registered grandchild branch point of the registered branch point data, and the matching child branch point is within the vicinity of the registration child branch point. Sometimes, the step of scoring the mismatch frequency based on the norm of the collator branch point and the registrant branch point, and when the matching grandchild branch point is in the vicinity of the registered grandchild branch point, the matching grandchild branch point and the registered grandchild Scoring disagreement frequency based on norm with bifurcation point, Calculating the average disagreement frequency for the branch point candidates, selecting the one having the smallest average disagreement frequency among the average disagreement frequencies of the parent branch point candidates as the parent branch point of the matching data, and the matching parent branch When the average disagreement frequency of points is less than or equal to a predetermined value, the step of determining the person as the true person is provided.

According to a sixteenth aspect of the present invention, in the process of registering the fingerprint, a step of registering feature points, a step of matching feature points of matching image data and registered data, and a step of registering template matching data. , A step of collating the template of the collation image data and the registered data, and a step of selecting one of the feature point collation and the template collation as the collation method of the fingerprint depending on the level of recognition rate of each fingerprint image. There is.

[0022]

The operation of the means of the present invention is as follows. In the present invention, first, the thinning processing unit determines a deletable point by a table search in units of 8 neighboring cells, and also executes scan control for reducing the number of scans. Then, the feature point extraction means extracts a feature point candidate from the image data subjected to the thinning processing by a table search in units of 8 neighboring cells, and deletes the pseudo feature point from the feature point candidate. Further, the characteristic point of the collated fingerprint image data and the registered characteristic point are collated by the characteristic point collating means, and the collated fingerprint image data and the registered data are template collated by the template collating means. Therefore, collation can be performed quickly while maintaining a high recognition rate, and a low cost and small fingerprint collation device can be realized.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. 1 to 12 are views showing an embodiment of a fingerprint collation device and a fingerprint collation method according to the present invention, and an example in which a collation fingerprint image data is applied to a registered fingerprint image data to judge in real time. Is. First, the configuration will be described. FIG. 1 is a block diagram of the fingerprint collation device 10. In this figure, the fingerprint collation device 10 includes a fingerprint image input device 11 and an A
/ D conversion device 12, frame memory 13 for storing and storing image data, image arithmetic processing device 20, read-only memory 21 for storing adjacent point pattern index values used for thinning scanning, and characteristic points A read-only memory 22 used for extraction scanning, a memory 23 for storing control information for thinning scanning, and a memory 41 for storing registration data.
, An input device 42, an output device 43, an access control signal generation device 44, and a central processing unit 40.

The image input device 11 comprises a transparent glass plate on which a finger to be inspected is placed, a light source for uniformly irradiating a fingerprint, a lens for focusing reflected light, and a CCD area sensor. . The video output signal of the CCD area sensor is sampled through the A / D converter 12, and the output digital code is transferred and stored in the frame memory 13. The image calculation processing device 20 includes a DSP (Digital) that executes high-speed product-sum calculation and vector calculation.
a read-only memory device 21 in which a signal processor (digital signal processor: digital signal processor) is stored, and an adjacent point pattern index for table search for thinning processing is stored, and an adjacent point pattern index for table search for feature points is stored. It is connected to the read-only memory device 22 by a bus to perform high-speed image processing calculation. The input device 42 is a device for inputting a command for changing / deleting a password and registration data in the registration / verification process. The output device 43 is a device that outputs the determination result in a human-readable format, and may be, for example, a display device, a voice output device, or a printer. The central processing unit 40 includes at least the input / output device 4
2, 43, access control device 44, management of image data,
Performs history management processing during the registration / verification process. The access control device 44 is configured to include a control signal generation circuit that generates, for example, a control signal for opening and closing a door based on the fingerprint collation result.

Next, the operation of this embodiment will be described. Thinning processing First, the thinning processing means of the fingerprint collation device 10 executed in the above hardware configuration will be described in detail (see FIGS. 2 to 4). Thinning is a process of extracting line information from a block of "1" of image data. Specifically, when the value of the pixel of a certain point of interest is "1", the pixel of the pixel adjacent to this point of interest is The value of the pixel at the point of interest is converted from "1" to "0" according to the pattern, and the screen data is sequentially and repeatedly scanned,
When there are no more cells in the screen data that change from "1" to "0", it is determined that the thinning process has converged, and scanning is stopped.

According to the fingerprint collation device 10, in the 8-neighboring cell shown in FIG. 2, that is, in the matrix of 3 pixels × 3 pixels, eight adjacent short pixels P _i adjacent to the center point P _o are used.
In (i = 1 to 8), a power of 2 which is a value corresponding to each bit of bit 0 to bit 7 of 1-byte data 2 ⁱ⁻¹ is given as a weighting coefficient W _i . A pixel value X _i (i = 1 to 8) of each adjacent point P _i is assigned a weighting factor W _i (i =
The adjacent dot pattern index value related to the center point _{P o} the sum ^{_{V p = Σ 8 i = 1}} X i × W i obtained by adding about 8 contiguous point value multiplied by 1-8). The pixel value X _i of each of the adjacent points
Is a binarized value of “1” or “0” and the weighting factor assignment method, the value V _p of the adjacent point pattern index is when all bits from bit 0 to bit 7 are off. 0 to 25 when all bits are on
The value is up to 5, and is stored in 1 byte.

The adjacent point pattern index is calculated for all combinations of which the central point can be changed from "1" to "0" in 8 neighboring cells of a 3 × 3 matrix, and the adjacent point pattern index values are calculated. The memory device 2 in advance
1 is stored and accumulated as a thinning table (FIG. 3).

When the pixel value of the center point of the 3 × 3 matrix is '1', the pixel value of the center point is changed from '1' to '0' without changing the connectivity of the figure. The thinning determination processing for determining whether or not the data can be deleted from is performed as follows. The adjacent point pattern index V _p for the eight adjacent points of the center point is calculated, and a table search is performed to see if the same value as V _p is in the thinning table. When the same value as V _p exists in the thinning table, the value of the center point is changed from '1' to '0', and when the same value as V _p is not in the table, the value of the center point is Hold it as '1'.

When there is no value in the thinning table that matches the neighboring point pattern index of the 8 neighboring cells, it is said that the thinning of the neighboring cell has converged. It is said that when the neighboring point pattern index of the neighboring cell exists in the table and the pixel value of the center point of the neighboring cell is changed from "1" to "0", the thinning of the neighboring cell has not yet converged. The scanning control means for controlling the thinning scanning of the screen obtained by cutting out a predetermined area of the input image data sequentially scans the screen while moving the center point of the 8 neighboring cells pixel by pixel, and sequentially scans the neighboring cells in units of Repeat until there are no pixel points that can be deleted on the screen.

FIG. 4 is a flow chart showing the thinning process. In FIG. 4, the symbols Sn (n = 1, 2, ...) Show each step of the flow. First, when the thinning scanning is started in step S1, the screen scanning is initialized in step S2, and in step S3, the process for the number of screen lines or less is repeated as the first stage screen scanning. That is, step S
In 3A, the thinning processing of 8 neighboring cells is performed, and step S3B is performed.
It is determined whether or not the thinning of the eight neighboring cells has converged. 8
If the thinning of the neighboring cells has not converged, step S3C
Then, the thinning convergence flag is turned on, position information is registered, and it is determined in step S3D whether or not movement of one pixel has been completed and processing of one line is completed. When the thinning of the 8 neighboring cells has converged, the process proceeds to step S3D as it is, and if the processing of one line is not completed in step S3D, the process returns to step S3A until the processing of one line is completed. In step S4, it is determined whether or not all the thinning convergence flags of the screen lines are off. If all the thinning convergence flags are off, the process is ended. If not all of them are off, the screen after the second stage is displayed in step S5. Start scanning. Next, in step S6, the scanning start point is skipped to a line in which the thinning convergence flag of the screen is ON,
In step S7, the scan is skipped over the registered neighboring cells. Next, in step S8, the 8-neighbor thinning process is performed for each pixel, and in step S9, the thinning flag and the pixel position are updated. Next, in step S10, it is determined whether or not all the thinning convergence flags of the lines on the screen are off. If not all of the thinning convergence flags are off, then in step S11 the predetermined number of screen scanning iterations is exceeded. Determine whether or not. When the predetermined value of the number of screen scanning repetitions is exceeded, or when the thinning convergence flags are all off in step S10, the processing of this flow is ended, and when the predetermined value of the number of screen scanning repetitions is not exceeded. Returning to step S5, the above process is repeated.

As described above, according to the fingerprint collation apparatus 10, after resetting the scanning control information for thinning, at least in the first screen scanning, the center points of the eight neighboring cells are screened in pixel units. From the upper left to the lower right-The scanning of the thinning process is sequentially performed in line units. For each line in this screen scanning, a thinning convergence flag that is turned on when there is a neighboring cell in which thinning does not converge in the line, and position information in the line of the neighboring cell in which thinning does not converge, The scanning control information for thinning consisting of the number of pixel points for which continuous thinning does not converge is calculated and stored in the memory.

When the scanning of the screen data is completed and the thinning convergence flag of all the scanned lines is off, it is judged that the thinning processing of the screen has converged and the scanning of the thinning processing is stopped. When at least one thinning convergence flag is ON in the scan target line, the thinning scanning of the screen data is repeated again. In this iterative process, the scan control information stored in the previous screen scan is used to move the scan line to the line where the thinning convergence flag is turned on, and the thinning on the scan line is reduced. From the first pixel point that does not converge, thinning scanning in units of neighboring cells is started, scanning is performed pixel by pixel on the line to determine that deletion is possible, and the scan control information of the line is updated. That is, when the thinning of the cells in the vicinity of the scanned line converges, the convergence flag of the line is reset to off. When there are continuous points on the line where the thinning has not converged, the thinning scanning on the line is skipped by the continuous number of points. The thinning scanning of the screen is repeated with the convergence flag of all lines turned off or repeated a predetermined number of times.

Feature Point Extraction Process Next, the means for extracting feature points by the fingerprint collation apparatus 10 will be described in detail (see FIGS. 5 to 7). Among the characteristic points, the adjacent point pattern indexes of the eight neighboring cells that display the patterns of intersections, branch points, and end points are stored in advance in the lookup table. A predetermined area of the image data that has undergone the thinning process is calculated in units of 8 neighboring cells, and the adjacent point pattern index is calculated.
Perform a table search in the table. When the same value exists in the table, the position coordinates of each of the feature points of the intersection, the branch point, and the end point are stored in the memory as a complement of the feature point. For example,
It is determined that the neighboring point pattern index value is 170 or 85 when the pixel value of the center point of the eight neighboring cells is "1".

When a pair of feature points of the same type are close to each other, the feature point pair is judged to be a pseudo feature point due to noise and is deleted from the memory to extract a true feature point. The respective directions of the feature point pairs of the same type are determined, and when the directions of the feature point pairs match and the distance between the feature point pairs is equal to or less than a predetermined value, it is determined to be a pseudo feature point. FIG. 7A is a pseudo-intersection when the distance between the intersections is short, and is removed from the intersection feature point candidates as a bridge. FIG. 7B is a pseudo branch point in which a branch point is generated by sweat, for example, and the pair of branch points is removed from the candidate branch point feature points. FIG. 7C is an end point pair in which the distance between the end point pairs, which is a feature point, is extremely small, and the end point pair is removed from the end point feature point candidates on the assumption that noise, which is a break, occurs in the originally connected line figure.

Feature Point Registration Processing According to the fingerprint collation apparatus 10, the feature point registration means first selects the branch point closest to the center point of the scanning area of the image data as the parent origin M1. Next, using the parent origin as the origin of the XY axis coordinates, the child branch point m that is the shortest branch point from the parent origin in the first to fourth quadrants.
_i (i = 1 to 4) is selected. Next, grandchildren branch point m _ij four branch point without overlapping from those distances from the child branch point in each quadrant is less mutually sequentially
(I, j = 1 to 4), and a maximum of 21 branch points are selected (see FIG. 8).

The feature point registration data includes the total number of branch points and their position addresses, relational data indicating the mutual positional relationship between the parent branch point M1 and the child branch point m _i in polar coordinates, and the child branch point m _i . Relational of grandchild branch point m _ij
It consists of data and. Relational data in the parent branch points M1 and child branch point _{m i} is a distance _{r i} of M1 and _{m i,} 9
The direction of the m _i and polar coordinates expressed in a counterclockwise direction at an angle theta _i respect ridge direction of M1 indicated by the arrow. However, the ridge direction of M1 is determined by a well-known method using binarized image data for each processing region by a direction filter in eight directions. The relative positional relationship between the child branch point and the grandchild branch point is expressed in polar coordinates by the distance θ between the child branch point and the grandchild branch point and the angle θ based on the ridge direction of the parent branch point.

Collation Processing Next, the collation processing of the feature points of the collated fingerprint image data and the registered feature points will be described in detail below. FIG. 10 is a flowchart showing the matching process of the feature points of the matching fingerprint image data and the registered feature points.

First, when the matching of feature points is started in S20, it is determined in step S21 whether the difference between the total number of branch points between the matching data and the registered data is less than a predetermined value, and the branch points are checked. If the difference between the total numbers is not less than the predetermined value, step S
In step 22, it is determined to be another person, and the processing of this flow ends. Also,
When the difference in the total number of branch points is less than or equal to a predetermined value, it is determined that the person is the person, and the parent branch point candidate of the matching data is selected in step S23, and the parent branch point candidate is set as the XY coordinate origin in step S24. Next, in step S25, four checker branch points are selected in the first to fourth quadrants, and in step S26, three or more checker branch points are respectively within a predetermined neighborhood centered on the registered branch point. Determine whether or not. When three or more collator branch points are within the predetermined vicinity centered on the registered branch point, the degree of disagreement based on the norm between the child branch point and the registered child branch point is calculated in step S27, and step S2
In step 8, a grandchild branch point is selected for each child branch point, and in step S29, it is determined whether or not three or more matching grandchild branch points are within a predetermined vicinity centered on the registered branch point. When three or more matching grandchild branch points are within a predetermined vicinity centered on the registered branch point, the degree of disagreement based on the norm between the grandchild branch point and the registered grandchild branch point is calculated in step S30, and the average disagreement is obtained in step S31. Calculate frequency and step S3
Go to 2.

On the other hand, in step S26, when three or more matching child branch points are not within the predetermined neighborhood around the registered branch point, the average mismatch frequency of the child branch points is set to 100 in step S33, and the process proceeds to step S32. When the three or more matching grandchild branch points are not within the predetermined neighborhood around the registered branch point in step S29, step S29 is performed.
At 34, the average degree of disagreement at the grandchild branch point is set to 120, and the process proceeds to step S31.

In step S32, it is determined whether or not the number of selected parent branch point candidates exceeds a predetermined value, and when the number of selected parent branch point candidates exceeds a predetermined value, step S3
In step 5, when the number of selected parent branch point candidates does not exceed the predetermined value, the process returns to step S23 to repeat the above process. In step S35, the matching parent branch point having the smallest average mismatch frequency is selected, and in step S36, it is determined whether or not the average mismatch frequency of the parent branch point is less than or equal to a predetermined value. When the average disagreement frequency of the parent branch points is less than or equal to a predetermined value, the matching feature point is determined to be the person in step S37, and when the average disagreement frequency of the parent branch points is not less than or equal to the predetermined value, it is determined to be another person in step S38. The process of this flow ends.

According to the above flow, the total number of branch points, the position address, and the relational data are extracted from the collation data in the same manner as the registration data. It is confirmed that the absolute value of the difference between the total number of extracted branch points and the total number of branch points of the registered data selected as the matching candidate is less than the predetermined value TH1. When the absolute value of the difference between the branch points is TH1 or more, it is determined that there is no identity between the collation data and the registered data, that is, it is determined as another person.

When the absolute value of the difference between the branch points is less than TH1, the branch point closest to the center in the matching data is selected as the parent branch point candidate M'1 and the matching parent branch point candidate M'1.
Assuming the parent branch point M1 registration branch point, carried out as follows determined whether the child branch point m ₁ in the matching branch _{point, m 2, m 3, m} 4 are present. Parent branch point candidate M'1
Of the child branch points m _i (i =
It is determined whether or not a matching branch point exists in a predetermined area section centered on the polar coordinate display (r _i , θ _i ) of 1 to 4).

[0043] Specifically, there acceptable vicinity _{(r i ± C ri, θ} i ± Cθ i) centered on the child branch point _{m i} of the registration data the child branch point _{m i} collation branch points in the region of Check whether to do or not. C _ri represents the permissible section width of the polar coordinate distance, and C θ _i represents the permissible section width of the polar coordinate angle. When there are two or more child branch points that do not fall within the area section among the four child branch points m ′ _i of the matching branch points, the parent branch point candidate M′1 is determined to be ineligible, and will be described below. The average value for the child branch points of the mismatch frequency is set to 100, and the process proceeds to the next parent branch point candidate process. When three or more child branch points are in each of the above neighboring areas, the degree of disagreement is scored. According to the fingerprint collation device 10, the child branch point m _i (i = 1 to 1)
The disagreement frequency S _pi of 4) is expressed by the following equation using the norm concerning the distance r and the angle θ in polar coordinate display. However,
(R _x , θ _x ) is a polar coordinate display of the child branch point m ′ _i with respect to the parent branch point candidate of the collation image data.

[0044]

[Equation 1]

The child branch point of the collation image data m _'i
For grandchild branch point _{m 'ij (i, j =} 1~4) are also in the same manner as the discrepancy degree _{S pi} of the verification data and the registered data of the parent branch point and said child branch point regarding _{the S pi,} polar coordinate It is expressed by the following equation using the norm concerning the distance r and the angle θ. (R _xx , θ _xx ) is the polar coordinate of the relative position of the verification grandchild branch point m ′ _ij with respect to the child branch point m ′ _i . (R
_ij and θ _ij ) are polar coordinates of the relative position of the grandchild branch point m _ij of the registered data with respect to the child branch point m _i .

[0046]

[Equation 2]

The weight of the mismatch frequency S _pij of the grandchild branch point m ′ _ij (ij = 1 to 4) is made 20% larger than the mismatch frequency S _{pi of the} child branch point m ′ _i (i = 1 to 4). This is because the emphasis is placed on the relationship that when the child branch point m ′ ₁ of the matching data matches the registered child branch point m ₁ , the grandchild branch point m ′ _j of the matching data matches the registered grandchild branch point.

In each matching of the matching branch point with respect to the parent branch point M1 and the child branch point m _i, and the child branch point m _i and the grandchild branch point m _ij, and the branch point in the registered data, the matching branch point is the above-mentioned each. When it does not exist in the area section, the mismatch degree of the child branch point of the matching data is set to 100, and the mismatch degree of the grandchild branch point of the matching data is set to 120. However, among the child branch points m ′ _i (i = 1 to 4) of the branch points of the matching data, the child branch points that are in the first quadrant to the fourth quadrant with respect to the parent branch point M′1 of the matching data. m ′ _i (i = 1 to 1
When the condition of 4) is not satisfied, the child branch point m ′ _i is not included in the calculation of the average degree of disagreement. Further, when the grandchild branch point m ′ _{ij of} the matching data does not exist in the quadrant in which the child branch point m ′ _i exists, the grandchild branch point m ′ _ij is not included in the calculation of the average mismatch frequency. Each child branch point m ′ _i (i =
1 to 4), the grandchild branch point m ′ that does not fall within the area section
When there are two or more _ij (i, j = 1 to 4), the average degree of disagreement is 120.

Parent branch point candidate M'1 of branch point of collation data
, The parent branch point is a value obtained by dividing the sum of the mismatch degrees of the child branch points m ′ _i (i = 1 to 4) and the grandchild branch points m ′ _ij (i, j = 1 to 4) by the number of branch points. The average mismatch frequency of the candidate M′1 is set.

From the branch point near the center of the screen data,
A predetermined number of parent branch point candidates M′1 are selected, the average degree of mismatch of each parent branch point candidate M′1 is calculated, and the parent branch point candidate having the smallest average degree of mismatch is calculated as the matching data. The parent branch point is M'1. However, the child branch point m ′ _i (i = 1, i = 1) of the grandchild branch point m ′ _ij (i, j = 1 to 4) is changed only when the average failure frequency of the parent branch point candidate M ′ 1 is corrected downward.
Calculate the average disagreement frequency for 4). By adopting this method, the processing time is shortened. Then, when the value obtained by dividing the sum of the mismatch frequencies of the child branch points m ′ _i (i = 1 to 4) of the parent branch points having the smallest average mismatch frequency by the number of child branch points is equal to or less than a predetermined threshold value, It is determined that the characteristic points of the collation data and the registration data are the same, and the person is recognized as the person.

According to the fingerprint collation apparatus 10, as shown in FIG. 11, registration of data for template collation is performed with a reference area A of 128 × 128 pixels with the center of the binarized image data as the center point. 64x1 on both left and right sides of the reference area A
The reference areas B and C of 28 pixels are taken, and image data obtained by compressing these reference areas to 1/4 is used as registration data. 1 /
In the compression of 4, the pixel value is "1" with 4 pixels as one cell.
When the number of cells is two or more, the cell is set to "1", and when the number of cells is less than two, the cell is set to "0". Next, with 48 × 48 pixels as one section, the number of pixel values “1” for each 8 × 8 = 64 sections is obtained and registered as gradation distribution data.

In the template matching of the binarized data, first, the X-axis and the Y-axis of the matching screen and the registration screen are aligned. For alignment, the shift (K, L) is obtained from 48 × 48 pixels as one section and the gradation distribution data of “1” in 8 × 8 = 64 sections. The deviation (K, L) is
The number of pixel values '1' of 48 × 48 pixels of the matching data and the registration data is obtained, and A (i, j) and B (i,
j) (i, j = 1 to 8). Deviation S (K, L)
(K, L = −7 to 0 to 7) is calculated by the following equation, for example, 2
55 deviations are calculated.

[0053]

[Equation 3]

[0054]

[Equation 4]

The deviation (K, L) that gives the smallest deviation is obtained. Focusing on the deviation (K, L), first, the collation screen A ′ and the registration screen A are collated every 5 pixels, and the degree of difference represented by the number 1 of the exclusive OR of the collation screen and the registration screen is minimum. Find the position coordinates of. Next, the difference between the matching screen and the registration screen is calculated on a pixel-by-pixel basis centering on the minimum position coordinate, and the position where the difference is the minimum is determined as the deviation.

The reference screens A'and A are compared in parallel by moving the collation screen by the above-mentioned displacement, and when the difference degree is smaller than a predetermined value TH1, the person is the original person, and when the difference degree is larger than the predetermined value TH2, the other person is selected. To determine. The degree of difference is a predetermined value TH1 and TH
2 between the reference areas B and C in FIG. 11, the reference areas have different degrees of difference TH3 and T, respectively.
If it is H4 or less, it is determined to be the person, and otherwise it is determined to be another person.

A specific embodiment will be described below with reference to the flow of the fingerprint registration process of FIG. First, the fingerprint registration process will be described. Fingerprint image data of an area of about 2 × 2 cm of the finger to be inspected placed on the image processing device is photoelectrically converted by the CCD area sensor (not shown) of the image input device, and the A / D converter 12 is set. Through the frame memory 13
Stored and accumulated in.

The image data of the fingerprint stored in the frame memory is subjected to preprocessing such as smoothing. For smoothing (smoothing) of image data, noise removal is performed by a well-known averaging filter. Next, the image data is binarized. A well-known method is used as the threshold value interpolation method depending on the screen area of the binarization process.

Further, thinning processing of the binarized image is performed,
The feature points of the thinned image data are extracted, and the above-mentioned data relating to the branch points are registered as feature point registration data (step S41). Next, in step S42, the collation data is collated by the feature point collation method based on the image data (step S42), and the template collation data for the binarized image data is registered in the memory (step S43). Here, the well-known eight-direction filter is used to determine the ridge direction of the fingerprint for template matching. Template matching is performed using the image data in the frame memory as matching data (step S44).

Then, in step S45, the recognition rates of the feature point matching method and the template matching method are compared. When the template matching method has a better recognition rate, the template matching is determined as the matching method of the finger to be inspected. If adopted and the feature point matching provides a better recognition rate, the feature point matching method is adopted and the registered image data and the matching method are associated and stored in the memory (step S46). Specifically, the template matching method is selected when the feature point matching method cannot distinguish the same fingerprint as the person. As the recognition rate of the matching method, the feature point matching method uses the ratio of the average disagreement frequency and a predetermined value for determining the true person, and the template matching method uses the ratio of the difference degree and the predetermined value for determining the right person. Can be quantified.

Note that the template matching and the feature point matching method selection in the fingerprint registration process are performed based on one screen scan data. It may be stored in the frame memory. At this time, the screen scanning time is about 50 milliseconds, and it is not necessary to place the finger on the fingerprint collation device many times, and the collation data can be scanned and stored during one imprinting.

In the registration process, the ID number or password is input and used as the index of the registered fingerprint data. In the collation process, first, the ID number or password is input, and then the finger to be inspected is placed on the image input device. The ID number is used as an index key to access the registered fingerprint data in the matching process to speed up the search of the registered data. The ID number selected in the registration process may be, for example, the same value for each group of a plurality of persons. In this case, in the verification process, the verification data is sequentially calculated for a plurality of registered fingerprint data having the same ID number. Compare and collate.

The image signal preprocessing, binarization processing, thinning processing, feature point extraction processing, and feature point matching image processing described above are all performed by the digital signal processor. The calculation of the index value of the adjacent point pattern of the eight adjacent points in the thinning processing is performed at high speed by a digital signal processor adopting a design method suitable for product-sum calculation. As another calculation method, a register having a bit width of at least 1 byte is initially cleared, and the pixel value of each adjacent point is set to "1".
At this time, the operation is also performed at high speed by sequentially scanning 8 adjacent points and turning on the bit in the byte corresponding to each adjacent point.

The thinning determination means and the table search for feature point extraction are
Specifically, although the memory is linearly searched by a digital signal processor, a binary search is performed or the value of the adjacent point pattern index is read in 1 byte (8 bits) by hardware. In a device equipped with means for accessing the read-only memory as a read address, read-only read-only
It may be performed by a means for determining whether or not to execute the thinning processing according to the memory content. The calculation time until the thinning processing of the screen of about 500 × 500 pixels by the lookup table operation method of the present invention converges is about 3
It is shortened to 0 to 50 milliseconds. Moreover, the response time of the collation of the present invention is about 1 second.

[0065]

As is apparent from the above description, the fingerprint collation apparatus of the present invention has a thin line processing and a characteristic point by the configuration in which the digital signal processing apparatus suitable for the image processing operation is provided independently of the central processing unit. High-level recognition that cannot be achieved by the feature point extraction method alone is performed by performing the extraction processing at high speed in units of 8 neighboring cells and providing a means for making a collation determination with a high recognition rate by the template matching method for fingerprints lacking salient feature points. It maintains the rate and provides a short-term verification response of the order of 1 second. Furthermore, the combination of the digital signal processor and the microprocessor provides an economical fingerprint matching device that is more cost-effective than the configuration of a single high-performance microprocessor.

[Brief description of drawings]

FIG. 1 is a hardware configuration diagram of a fingerprint matching device.

FIG. 2 is a schematic diagram showing a positional relationship between adjacent points of eight neighboring cells.

FIG. 3 is a table of adjacent point pattern index values in which a pixel at a center point of 8 neighboring cells can be deleted.

FIG. 4 is a flowchart showing thinning processing.

FIG. 5 is a table of adjacent point pattern index values of 8 neighboring cells at intersections, branch points, and end points.

FIG. 6 is an example of a pattern of 8 neighboring cells at intersections, branch points, and end points of feature points.

FIG. 7 is an example of pseudo feature points at intersections, branch points, and end points.

FIG. 8 is a diagram exemplifying a mutual relationship among a parent branch origin point, a child branch point, and a grandchild branch point of a registered branch point (here, a mark represents a registered branch point, and an X mark represents another branch point. Represent).

9 is a partially enlarged view of FIG. 8 showing a polar coordinate system of relations of registered branch points.

FIG. 10 is a flowchart showing a process of a feature matching unit.

FIG. 11 is a schematic diagram showing a reference region for template matching.

FIG. 12 is a flowchart showing a fingerprint registration process of the fingerprint matching device.

[Explanation of symbols]

11 Fingerprint image input device 12 A / D conversion device 13 Frame memory 20 Image processing unit 21 Read-only storage for adjacent point pattern index value
Memory 22 Read-only memory used for feature point extraction 23 Thinning scanning control information storage memory 40 Central processing unit 41 Registered data storage memory 42 Input device 43 Output device 44 Access control signal generation device

Claims (16)

[Claims] 1. A fingerprint collation device that collates collated fingerprint image data with registered fingerprint image data and suppresses selection of access or access based on a determination result, which is a binary value in which pixel values are binarized into white and black. A thinning processing unit for thinning the thinned image data, the thinning determining unit determining a deletable point by a table search in units of 8 neighboring cells, and a scanning control for reducing the number of scans of the thinning determining unit. A thinning processing unit including means, and a feature point extracting unit for extracting a feature point of image data, wherein feature point candidates are extracted from the thinned image data by a table search in units of 8 neighboring cells. Means, and pseudo feature point removing means for deleting pseudo feature points from the feature point candidates,
Feature point extraction means, means for registering true feature points of image data, feature point collation means for determining the feature points of collated fingerprint image data and registered feature points, and template data for template collation And a template matching unit for matching the matching fingerprint image data with the registered data as a template. 2. The thinning determination means, in an eight-neighbor cell consisting of a center point and eight adjacent points,
Calculate an adjacent point pattern index value that is the sum of the product of a predetermined weighting coefficient pre-assigned to the position of each adjacent point with respect to the center point and the pixel value of that adjacent point for each of the eight adjacent points. And a step of performing a table search in the memory device storing the respective adjacent point pattern index values for all the combination patterns in which the pixel value of the center point of the 8 neighboring cells can be changed from '1' to '0'. 2. When the adjacent point pattern index value exists in the table, the step of changing the pixel value "1" of the center point of the eight neighboring cells to "0" is included. Fingerprint matching device. 3. The screen scanning in the first step in which the scanning control means scans a screen of a predetermined size of image data by the thinning determination means, and scans all the lines of the screen for each line for each pixel. A process for applying the thinning determination means, at least a flag indicating whether or not the thinning of the scanning line has converged, and position information in the line of the neighboring cells in which the thinning in the line has not converged, A screen scanning process in a first step, which includes a process of storing scan control data including a number of neighboring cells in which the thinning in the line has not converged continuously, and a screen in the second and subsequent stages. The scanning process, which is selected only when the thinning process of the screen scanning in the previous stage of the stage does not converge, and based on the scan control data in the previous stage of the stage, the thinning determination process is performed. Scan the And the screen scanning processing reduction of the total number of pixels and a process of updating the contents of the scan control data, and a second
The screen scanning processing after the step and the convergence determination step for determining whether or not the screen thinning has converged. When the screen thinning does not converge, the screen scanning processing from the second step is repeated. The fingerprint collation apparatus according to claim 1, further comprising: a repetitive convergence determination step. 4. The means for extracting the feature points: a step of calculating an adjacent point pattern index value of eight neighboring cells; and an adjacent pattern index value is stored as an adjacent point pattern index value of an intersection, a branch point and an end point. And a step of registering the neighboring cell as a feature point candidate when the same adjacent store pattern value exists in the memory. Item 1. The fingerprint collation device according to item 1. 5. The pseudo feature point removing means selects from the feature point candidates when the pairs of feature point candidate pairs of the same type have the same direction and the distance between the feature point candidate pairs is a predetermined value or less. The fingerprint collation device according to claim 1, wherein the fingerprint collation device is deleted. 6. The step of registering the feature point, selecting a branch point closest to the center point of the screen data as a parent branch point, and using the parent branch point as the origin of the XY coordinate axes, the first quadrant, In each of the two quadrants, the third quadrant, and the fourth quadrant, the step of selecting the branch point closest to the parent branch point as the child branch point of each quadrant, and the child branch in each quadrant Selecting a predetermined number of branch points as grandchild branch points in the order of proximity to the branch points, at least the total number of branch points, position information of the branch points, and mutual positional relationship between the parent branch point and the child branch point, The fingerprint collation apparatus according to claim 1, further comprising a step of registering a mutual positional relationship between the child branch point and the grandchild branch point. 7. The step of comparing the feature points with a step of comparing the total number of branch points of the collation data with the total number of branch points of the registered data, and the branch points as parent branch point candidates in the order of proximity to the center point of the collation data. A step of selecting a predetermined number of steps, a step of determining whether or not the child branch point of the parent branch point is within the vicinity of the corresponding child branch point of the registered branch point data, and a step of collating the child check branch point Determining whether the grandchild branch point is within the vicinity of the corresponding registered grandchild branch point of the registered branch point data, and when the collator branch point is within the vicinity of the register child branch point, the collator branch point The step of scoring the mismatch frequency based on the norm of the point and the registrant branch point, and when the matching grandchild branch point is in the vicinity of the registered grandchild branch point, the norm of the matching grandchild branch point and the registered grandchild branch point is set. Based on the stage of scoring disagreement frequency based on A step of calculating an average mismatch frequency, a step of selecting a parent branch point candidate having the smallest average mismatch frequency as a parent branch point of the matching data, and an average mismatch of the matching parent branch points. The fingerprint collation apparatus according to claim 1, further comprising a step of determining that the person is the person when the frequency is less than or equal to a predetermined value. 8. The fingerprint registration process includes: registering feature points; matching feature points between collation image data and registered data; registering template collation data; and collation image data. And a template of registered data, and a step of selecting either feature point matching or template matching as a fingerprint matching method depending on the recognition rate of each fingerprint image. The fingerprint matching device according to claim 1. 9. A fingerprint collation method for collating collated fingerprint image data with registered fingerprint image data and suppressing selection of access or access based on a determination result, which is a binary value in which pixel values are binarized into white and black. A thinning processing step for thinning the thinned image data, a thinning determining step for determining a deletable point by a table search in units of 8 neighboring cells, and a scanning control for reducing the number of scans in the thinning determining step. And a feature point extracting process for extracting feature points of image data, wherein feature point candidates are extracted from the thinned image data by a table search in units of 8 neighboring cells. A process, and a pseudo feature point removal process of deleting the pseudo feature point from the feature point candidate,
Feature point extraction step, a step of registering the true feature point of the image data, a feature point matching step of determining the feature point of the matching fingerprint image data and the registered feature point, and template data for template matching And a template matching step of matching the registered fingerprint image data with the registered data as a template. 10. The thinning determination step comprises: in an 8 neighboring cell consisting of a center point and 8 adjacent points,
Calculate an adjacent point pattern index value that is the sum of the product of a predetermined weighting coefficient pre-assigned to the position of each adjacent point with respect to the center point and the pixel value of that adjacent point for each of the eight adjacent points. And a step of performing a table search in the memory device storing the respective adjacent point pattern index values for all the combination patterns that can change the pixel value of the center point of the 8 neighboring cells from '1' to '0'. 10. When the adjacent point pattern index value is present in the table, the step of changing the pixel value "1" of the center point of the eight neighboring cells to "0" is included. Fingerprint matching method. 11. The scan control step is a first-step screen scan in which a screen of a predetermined size of image data is scanned by the thinning determination step, and all lines of the screen are scanned line by line and pixel by pixel. A process for applying the thinning determination means, at least a flag indicating whether or not the thinning of the scanning line has converged, and position information in the line of the neighboring cells in which the thinning in the line has not converged, A screen scanning process in a first step, which includes a process of storing scan control data including a number of neighboring cells in which the thinning in the line has not converged continuously, and a screen in the second and subsequent stages. The scanning process, which is selected only when the thinning process of the screen scanning in the previous stage of the stage does not converge, and based on the scan control data in the previous stage of the stage, the thinning determination process is performed. Scan the And the screen scanning processing reduction of the total number of emissions and the pixel, the process of updating the contents of the scan control data, and a second
The screen scanning processing after the step and the convergence determination step for determining whether or not the screen thinning has converged. When the screen thinning does not converge, the screen scanning processing from the second step is repeated. 10. The fingerprint matching method according to claim 9, further comprising: a repeating convergence determination step. 12. The step of extracting the feature points comprises the steps of: calculating adjacent-point pattern index values of 8 neighboring cells; and calculating the adjacent-pattern index values as the adjacent-point pattern index values of intersections, branch points, and end points. A table search in the stored memory device, and a step of registering the neighboring cell as a feature point candidate when the same adjacent store pattern value exists in the memory. The fingerprint matching method according to claim 9. 13. The pseudo feature point removing step extracts from the feature point candidates when the pairs of feature point candidates of the same type have the same direction and the distance between the feature point candidate pairs is a predetermined value or less. The fingerprint matching method according to claim 9, wherein the fingerprint matching method is deleted. 14. The step of registering the characteristic points includes the step of selecting a branch point closest to the center point of the screen data as a parent branch point, the parent branch point as the origin of the XY coordinate axes, and the second quadrant, In each of the two quadrants, the third quadrant, and the fourth quadrant, selecting the branch point closest to the parent branch point as the child branch point of each quadrant, and the child branch in each quadrant Selecting a predetermined number of branch points as descendant branch points in the order of proximity to the points, at least the total number of branch points, position information of the branch points, the mutual positional relationship between the parent branch point and the child branch point, and the child 10. The fingerprint matching method according to claim 9, further comprising a step of registering a mutual positional relationship between the branch point and the grandchild branch point. 15. The step of collating the characteristic points includes a step of comparing the total number of branch points in the collation data with a total number of branch points in the registered data, and the branch points as parent branch point candidates in the order of proximity to the center point of the collation data. A step of selecting a predetermined number, a step of judging whether or not the child branch point of the parent branch point is in the vicinity of a corresponding registrar branch point of the registered branch point data, and a step of collating the collator branch point Determining whether the grandchild branch point is within the vicinity of the corresponding registered grandchild branch point of the registered branch point data, and when the collator branch point is within the vicinity of the register child branch point, the collator branch point The step of scoring the mismatch frequency based on the norm of the point and the registrant branch point, and when the matching grandchild branch point is in the vicinity of the registered grandchild branch point, the norm of the matching grandchild branch point and the registered grandchild branch point is set. Based on the score of the disagreement frequency based on Calculating the average disagreement frequency, the step of selecting the average disagreement frequency of the parent branch point candidates that has the smallest average disagreement frequency as the parent branch point of the matching data, and the average of the matching parent branch points. 10. The fingerprint collation method according to claim 9, further comprising a step of determining that the person is a person when the disagreement frequency is less than or equal to a predetermined value. 16. The fingerprint registration process includes: registering feature points; matching feature points between matching image data and registered data; registering template matching data; and matching image data. It is characterized in that it comprises a step of collating a template with the registered data and a step of selecting either one of the characteristic short collation and the template collation as the collation method of the fingerprint depending on the recognition rate of each fingerprint image. The fingerprint matching method according to claim 9.