JPS63132386A - Fingerprint collating method - Google Patents

Abstract

PURPOSE:To prevent a check error changing the size of a protruding line due to a pressure and the pattern of the protruding line according to the direction of the pressure by binarizing a fingerprint picture and thinning, extracting the core lines of the protruding line parts and recessed parts and collating the number of the intersections of the core lines between a registered fingerprint and an inspected fingerprint. CONSTITUTION:The finger print picture signal of a finger F through a video camera 14 is converted to a digital signal of 4-8bie, a correction by a shading correction circuit 21 is applied, it is binarized in a binarization circuit 22 and stored in a picture memory. A fetched picture decision part 30, when the sum of the area of the binarized protruding lines in an area is within a constant range, stores it in a fingerprint picture memory 31. A core line extracting part 40 reads data from the fingerprint picture memory 31 to thin the protruding parts and the recessed parts of the fingerprint, extracts the core lines to have core line picture data and stores in a memory 41. An area (ten plate) required for the collation is selected from a core line picture and recorded in a fingerprint registering file. A collation decision part 50 executes a rough collation and a fine collation in order to obtain a high speed processing. The number of the intersections is counted by the use of the plate for the rough collation. The fine collation is carried out to a part below a constant value in the collating error.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to pattern matching by image data processing,
In particular, the present invention relates to a fingerprint matching method suitable for identity verification.

[Prior art]

As a conventional example of fingerprint matching, see Information Processing Society of Japan, Vol. 12.
5. No, 6, (1984, June issue P, 599-P, 6
There is an example described in 05).

As shown in Fig. 16, this system consists of two systems: a fingerprint input subsystem and a fingerprint verification subsystem. First, the fingerprint input subsystem will be explained. It is read as value data, and then a low-pass filter is applied in the flow direction of the fingerprint ridges to remove noise and improve image quality.
After filtering is performed in the normal direction using an anisotropic filter having properties of a bandpass filter, the data is converted into binary data by floating binarization. Here, floating binarization is a method in which the average value of pixels (for example, an 8×8 area) surrounding the processing pixel is used as the threshold for binarization.

The ridges of the binarized fingerprint are thinned (skeletonized) and converted into core lines (
After creating a skeleton pattern), feature points (end points, branch points) are extracted by template matching, and their types and positions are recorded.

The extracted feature points are recorded as true feature points after a correction process is performed to remove feature points due to erroneous detection. The content of the correction process is to remove end points caused by breaks in ridges and branch points caused by small protrusions on ridges. Specifically, the vicinity of the end points in the ridge direction is examined, and pairs are removed. If the desired end point exists, it is determined that the end point is due to a break in the ridge,
If an end point exists in the vicinity of a branch of a ridge, it is determined that the branch point and end point are caused by a small protrusion of the ridge.

The position of the feature points extracted and corrected in this way is described by the skin surface, with the center of the fingerprint as the origin and the direction of flow of the ridge below the center as the Y axis, as shown in FIG.

In this way, the feature points whose position types are described are further
As shown in Figure 7, relationships called relationships between feature points are investigated. Here, a relation is defined as a local skin surface system (with the ridge direction as the Y-axis) centered around a certain feature point (parent feature point), and one feature point closest to the vertex in each quadrant. A total of 4 child minutiae points are selected, and the number of ridges R, , which exist between these points and the parent minutiae point is
R.

This refers to R3 and R4, and is intended to be considered as features attached to the parent minutiae.

Finally, for one minutiae, the position (x.

y), direction d, relation (R1, R2, R3, Ra
) information can be obtained. In other words, one feature point is (xry
, d, R+, Rz, R:l, Ra), and one fingerprint is finally a collection of descriptions of these minutiae points (approximately 1
00 pieces).

Next, the fingerprint matching subsystem will be explained.

Verification in this subsystem measures the degree of agreement between the registered fingerprint and the data representing the minutiae of the fingerprint input for inspection. Specifically,
As shown in FIG. 18, this process consists of coarse matching in which the entire fingerprint is aligned, and fine matching in which the degree of matching is measured based on the data of paired minutiae after alignment.

Rough matching is performed by first comparing the registered fingerprint (file fingerprint) and the test fingerprint (
The process begins with determining the minutiae to be paired with the search fingerprint. Here, regarding the feature points, as shown in Fig. 19,
(2) The difference in direction between the parent feature points, (2) The difference in the position and direction relation of the four child feature points, and if they are within the allowable values, they are considered as "pair" candidates.

Next, the registered fingerprint and the test fingerprint are separated, for example, at an angle of △θ(△
θ=π/32), 15 types are changed and superimposed, and the difference between each angle t′ and the position of the candidate feature point △x=lx, -Xr 1 △y=1y* 'jt 1 The above points (ΔX, Δy) are calculated and plotted on the ΔX-Δy plane as shown in FIG.

If this is done, a phenomenon will occur where the plots will be concentrated around the position (ΔX, ΔY) on the Δx - Δy plane at a certain rotation angle θ, but at this time (ΔX.

ΔY)θ is the difference in position and angle between the registered fingerprint and the test fingerprint.

Precise matching is performed using the above position and angle differences (△X).

After correcting ΔY)e, examine the ``pair 1'' candidates again, and select the ``pair 1'' minutiae that are consistent with the fingerprint as a whole, selecting them in descending order of political change. Examining the position difference, direction difference, and ridge number difference for the ``pair 1'' minutiae selected in this way, and further examining the position difference, direction difference, and ridge number difference for child minutiae,
The degree of matching of the minutiae of the entire fingerprint is summed up and finally output as the degree of matching of the fingerprints.

By the way, although the conventional technology described above has certainly achieved highly accurate fingerprint verification, the processing content is complicated and requires a dedicated high-speed arithmetic processing unit and a large general-purpose computer to put it into practical use. required.

On the other hand, the current method of verifying the identity of the person used for room entry control and transaction management is generally using a personal identification number, but this is powerless if a third party learns the number, so A reliable method is desired, and for this reason, fingerprint matching may be applied.

However, in order to satisfy such demands, it is necessary to develop a simple fingerprint matching algorithm that can be processed in real time even with the processing power of a terminal device.

Therefore, as a simple fingerprint matching algorithm, pattern matching can be considered. Note that this pattern matching method takes, for example, the sum of the absolute values of the differences between the registered fingerprint image and the fingerprint image to be inspected, that is, the matching error, and when this is smaller than a certain threshold value, the two images, i.e. It is assumed that the fingerprints match.

However, in this method, if there is a positional or angular shift between the images, the matching error will not be reduced and an erroneous determination will occur. Therefore, in order to avoid this, the matching error is usually calculated by changing the angle and position of one image little by little, and the angle and position of the two images are calculated at the angle and position that give the minimum matching error.
If the positions match and the matching error is less than a threshold, it is determined that the two are the same.

[Problem that the invention seeks to solve]

Conventional methods using pattern matching have the advantage of being easy to speed up because the algorithm is simple and can be easily made into hardware, and can be used for fingerprint matching.
There are the following problems related to the method of reading fingerprint images.

That is, first, for fingerprint verification, it is necessary to capture an image of the fingerprint. A method commonly used for reading fingerprints is one that utilizes total internal reflection of a prism. In this method, as shown in Figure 21(a), when a finger is pressed against the total reflection surface of a rectangular prism, only the ridges come into contact with the glass surface, and the light that should be totally reflected is directed toward the finger. This takes advantage of the fact that since the light passes through and comes out, only the ridges appear black when viewed from the direction of total reflection, as shown in FIG. 6(b).

This method is often used because it can easily obtain fingerprint images with good contrast, but since fingerprints are rubber-like elastic bodies, the thickness of the ridges in contact with the glass surface changes depending on the strength of the pressing pressure. The thickness of the ridges in the obtained image changes. Furthermore, the fingerprint is deformed depending on the direction of the pressure applied, and as is done in normal pattern matching,
Merely matching the angle and position does not sufficiently reduce the matching error.

The reason for this will be explained below with reference to FIG.

In Fig. 22, (a) and (b) show the same fingerprints collected by changing the pressing pressure.
(a) is the one under normal pressure, and (b) is the one under strong pressure.

By the way, these (a) and (b) appear to be exactly the same, but if you actually superimpose the binarized images shown in (c) and (d) in the same figure, you will see (e) in the same figure. ), the ridges almost match in the area around the center, which is about 3 tree lengths, but the ridge positions do not match in the area outside of that, and
Overall, the matching error does not become smaller. As a result, it becomes necessary to set a high level of matching error in order to determine that fingerprints are the same, and conversely, the probability that different fingerprints will be determined to be the same increases. Furthermore, even in a portion where the patterns almost match near the center, since the thickness of the ridges differs, a matching error occurs in this portion as well, and complete matching (matching error is 0) cannot be obtained.

For reference, FIG. 22(f) shows an example of pattern matching using different fingerprints.

An object of the present invention is to provide a fingerprint matching method that can sufficiently address the problems of the conventional methods described above, easily and reliably obtain fingerprint matching by pattern matching, and that can be easily applied to identity verification. There is a particular thing.

[Means for solving problems]

In order to achieve this object, the present invention is as follows.

That is, first, the problems of the above conventional example can be summarized into the following two points.

(b) The thickness of the ridge changes due to pressure, causing a matching error.

(b) The ridge pattern changes depending on the direction of pressure, causing a matching error.

As mentioned above, in the conventional pattern matching method, the difference is directly taken between image data and the sum of the absolute values of the difference within the matching area is used as the matching error. The deformation directly appears as a matching error.

Therefore, in the present invention, as shown in FIG. 2(a), the fingerprint image is first binarized and then thinned, and then
), the core lines of the ridges or valleys are extracted, and the number of intersections between the core lines of the registered fingerprint trough and the core lines of the test fingerprint ridge is calculated. The above-mentioned problems (a) and (b) can be alleviated by using the number of intersections between the core wire and the core wire in the trough of the inspection fingerprint as the matching error.

[Effect]

The change in ridge width due to the magnitude of the pressing pressure mentioned in (a) above has no effect on matching errors due to the thinning, and the deformation of the ridge pattern (fingerprint) depending on the pressing direction mentioned in (b) above. By thinning the lines, as shown in FIG. 3, the pitch width of the ridges becomes an allowable width for positional deviation, and the influence on matching errors is eliminated. That is, deformation of the ridge pattern within the allowable width does not result in a matching error because no intersection occurs. On the other hand, if line thinning is not performed, the deformation will directly result in a shift between the ridge patterns, resulting in a matching error even if the amount of deformation is minute.

〔Example〕

Hereinafter, the fingerprint verification method according to the present invention will be explained in detail with reference to examples.

First, the principle of the present invention will be explained with reference to FIG.

In Fig. 4, (a) shows a fingerprint image taken under normal pressure, and (b) shows a fingerprint image taken under strong pressure, which is different from the conventional example in Fig. 19. (a) in case.

This corresponds to (b).

In the present invention, the fingerprint images shown in (a) and (b) first become core lines due to thinning of the troughs on the one hand, and on the other hand, as shown in (c) and (d) of the same figure. After the ridges are thinned to form core wires, pattern matching is performed as shown in FIG. 2(e).

Now, when comparing the case of the present invention shown in FIG. 4 and the case of the conventional example shown in FIG. 22, first, (6) in FIG.
In Fig. 4(e), the ridge width is slightly different even in the center where both patterns match, causing a matching error, whereas in Fig. 4(e), the lines are thinner, so the central bone No intersection appears in the center, and the matching error is zero at the center.

Furthermore, in the periphery of the fingerprint shown in Figure 4(e), there are no intersections up to the four ends of the core wires, and compared to the case in Figure 19, the range in which the matching error is small (nearly zero) is widened. It can be seen that the case of the present invention is more resistant to deformation as a whole. In addition,
FIG. 4 ( ) shows the case of different fingerprints for reference.

However, even in the case of the present invention shown in FIG. 4, an intersection occurs when reaching further to the outside, and the matching error also increases. This is the area (range) in which the matching error is calculated during matching.
This shows that it is necessary to limit the deformation to a small region with little deformation. When the matching area is limited in this way, the recognition rate is influenced by the size of the area.

In other words, when the matching area is made small, it is less susceptible to deformation, but the probability that partial patterns match even with different fingerprints increases, making it more likely to cause misrecognition.On the other hand, when the matching area is made large, , it becomes more susceptible to deformation, and even the same fingerprint can be easily recognized as a different fingerprint.

Therefore, in the following embodiments of the present invention, a verification method that is not easily influenced by such verification area settings is also provided. That is, if matching is simply determined as the sum of the absolute values of images or the sum of the number of intersections, the problem of setting the matching area as described above will arise in this case.

Therefore, in order to solve this problem, in the following example, the size of the area where the pattern is defeated is measured by the absolute value of the difference or the distribution of the intersection point as a matching method, and the size of this matching area is calculated. If the number of fingerprints exceeds a certain level, the fingerprints are determined to be the same.

To explain this with Figure 4(e), when the registered fingerprint image and the test fingerprint image are the same fingerprint, when they are aligned and superimposed, no intersection will occur in the center as shown in the figure, and the matching error will be almost zero. However, due to deformation of the fingerprint in the peripheral area, the positional relationship shifts and an intersection occurs.

This distribution is schematically shown in FIG. 5(a), and the area shown in FIG. 5(C) is the same as that in FIG.

It exhibits the pattern shown in (d). Next, when different fingerprint images are superimposed, the intersection points are dispersed in various parts regardless of the position, and no pattern can be seen as in the case of the same fingerprint. This state is shown in FIG. 5(b).

It is schematically shown in (e).

Therefore, by examining the distribution of intersection points, that is, the size of the matching area in the image, if the size of this area is above a certain value, it is the same fingerprint, and if it is below a certain value, it can be determined that it is a different fingerprint. This method is adopted in the following embodiments.

With this method, there is no need to predict the deformation in advance and determine the matching area, and a practical matching method can be provided. Although such a matching method has been described using the example of the number of intersections, it goes without saying that it can also be applied to cases where the absolute value of the difference between images is used as the evaluation standard.

Hereinafter, one embodiment of the present invention will be explained in detail.

FIG. 1 shows an embodiment of the present invention, which generally includes a fingerprint reading section 1,
The signal processing section 2) is composed of an intake determination section 3, a fiber extraction storage section 4, and a collation determination section 5, and these will be explained in sequence below.

First, the fingerprint reading section 1 includes a light source 10, a diffusion plate 11, a prism 12, an imaging lens 13, a video camera 14, and the like.

The light emitted from the light source 10 is made into uniformly diffused light by the diffuser plate 11 and then enters the prism 12.

The prism 12 is composed of a right-angle prism, and the light that enters the prism 12 is normally reflected by a 45° total reflection surface and reaches the camera 14. When pressed, light is no longer reflected in the areas where the ridges of the fingerprint touch, but instead passes through to the ridges. When this is viewed from the camera 14 side, the pattern of ridges appears black on the total reflection surface, as shown in FIG. 18(b). At this time, the back shadow is imaged with the diffuser plate 11 out of focus. Therefore, without the diffuser plate 11, the light source 10 would be directly imaged, which is not preferable as a background of the ridge. Note that a method using such a prism is herein referred to as a right-angle prism imaging method.

The fingerprint pattern contrasted by the prism 12 is then imaged onto the imaging surface (COD, etc.) 14a by the lens 13 of the camera 14, but at this time, the total reflection surface of the prism 12 is aligned with the optical axis. Since the lens 13 is tilted relative to the optical axis, the image forming surface of the lens 13 also appears to be tilted with respect to the optical axis.

Therefore, in this embodiment, by installing the imaging surface 14a in accordance with this inclination, it is possible to focus on the entire total reflection surface of the prism 12, that is, the surface on which the fingerprint is to be imaged. There is.

This will be explained in more detail with reference to FIG.

In addition, this FIG. 6 represents the imaging lens 13 with a straight line indicating its diametrical direction. In this figure, the light that passes through the center O of the total reflection surface of the prism 12 is directed to the lens at the focal point F. 13 and is imaged at O'. On the other hand, if we look at the light from a point P on the total reflection surface, for example, according to geometric optics, the light that passes from this point P through the center of the lens 13 will travel straight, but from P parallel to the optical axis. The light emitted from the lens is refracted into light that passes through the focal point F, and intersects the light that exits from point P and passes through the center of the lens at point P', forming an image of point P at this point. Similarly, point Q on the total reflection surface forms an image at point Q'.

In this way, the total reflection surface of the prism 13 forms an image on the plane passing through P'O''Q', and unlike in a normal optical system, the image forming surface is not perpendicular to the optical axis.

Therefore, if the imaging surface is simply provided perpendicular to the optical axis, the image will be blurred. Therefore, in the above embodiment, the imaging surface 14a is tilted so that it coincides with the imaging surface formed by the lens 13. It is set up.

However, in this embodiment, the total reflection surface of the prism 12 is tilted with respect to the optical axis, and as a result, the fingerprint to be imaged is viewed from an oblique direction, resulting in image distortion. , defenseless.

Therefore, an optical system that can eliminate the occurrence of this distortion will be explained with reference to FIG.

In the optical system shown in FIG. 7, the lens 13 is arranged parallel to the total reflection surface of the prism 12, that is, the optical axis of the lens 13 is arranged at right angles to the total reflection surface of the prism 12. In this case, since the total reflection surface of the prism 12 and the lens 130 surface become parallel, the image formation surface also becomes parallel to the total reflection surface of the prism, and as a result, image distortion is eliminated.

Figure 7 shows this in terms of geometric optics, where PL:LP
”.

It can be seen that the ratios of OL:LO'', QL, and :LQ' are all the same.

In reality, if the optical axis of the lens and the optical axis of the optical system are shifted in this way, light enters the lens obliquely, which increases various aberrations and makes it impractical. Therefore, by arranging the angle between the optical axis of the optical system and the optical axis of the lens between zero and the above angle, the above aberration is relatively small.
Furthermore, image distortion can be kept to an acceptable level. An example of such an optical system is shown in FIG. In one embodiment of the present invention, an optical system is configured as shown in FIG.
What is necessary is to install the imaging surface 14a.

Next, the signal processing section 2 includes an A/D converter 20, a shading correction circuit 21, a binarization circuit 22, and the like.

The fingerprint image signal read from the finger F by the video camera 14 is converted into a 4 to 8 bie digital signal by the A/D converter 20, then corrected by the shading correction circuit 21, and then binarized. After being binarized by the circuit 22, it is temporarily stored in an image memory (described later).

The shading correction here is performed by storing the brightness distribution of the fingerprint background in advance and taking the difference between the actually input fingerprint images from this value, thereby making the brightness distribution uniform, This is done to reduce errors during binarization.

Furthermore, in the binarization by the binarization circuit 22, the portion where the value of the above-mentioned shading-corrected image [(background brightness distribution) - (fingerprint image)] exceeds a certain threshold value is treated as a black pixel. This is done by setting I11 and setting the portion below the threshold value to '0'' as a white pixel.

In addition, as another embodiment, the floating binarization method is not limited to the above-described binarization after shading correction.

Subsequently, the capture determining section 3 includes a capture image determining section 30 and a fingerprint image memory 31.

When a finger F is pressed against the total reflection surface of the prism 12, the shading of the ridges in the captured fingerprint image changes depending on the pressure and the state of perspiration on the fingertip, and in extreme cases, the ridges may be interrupted or faded due to scratches. , conversely, short circuits of ridges due to whelks occur,
This may cause erroneous judgments.

Therefore, it is necessary to determine the quality of the captured image and input it into the fingerprint image memory.

The captured image determining unit 30 is for this purpose, and when the area of the binarized ridges in a predetermined area, that is, the total sum of black pixels, is within a certain range, a good image can be obtained. It serves to store the fingerprint image in the fingerprint image memory 31 as a fingerprint image. In other words, the number of black pixels is small in a dull state, and the number of black pixels is large in a dull state, so in order to capture a good image except for these cases, the upper limit of the allowable range in the captured image setting section 30 is set. and the lower limit. In this way, the configuration can be simpler than the conventional method of directly detecting the shading of image ridges.

Furthermore, the core wire extraction storage section 4 is composed of a core wire extraction section 40 and a core wire image memory 41.

The core wire extraction unit 40 reads the fingerprint image data from the fingerprint image memory 31, thins the ridges and valleys of the fingerprint as explained in FIGS. 2 to 4, extracts the core wires, and extracts the core wires. The image data is stored in the fiber image memory 41.

At this time, for the ridges of the fingerprint, the above-mentioned thinning process can be performed with the image data as it is, that is, with the black pixels set to 11", but when thinning the valleys, the black pixel IF For I 11 and white pixel II OII, If
I II → II Q It, II Q IT-F
It is necessary to thin the line by replacing it with f I It. The data thinned in this way is stored as a binary image in the cardiac line image memory 41, as described above.

On the other hand, this heart line image data is also used when registering a new fingerprint, but in this case, an area (template) necessary for verification is selected from the above heart line image and then registered as a fingerprint. It is recorded in a file (described later). Note that, as the area for this verification, it is common to extract an area that includes the center of the fingerprint, and the recognition results are also relatively good.

By the way, the core wire extraction here is as described above.
This is a countermeasure against deformation of fingerprints, but as a concrete method to implement this, HILIDIT
Various algorithms proposed by ICH), Murata et al. may be used. However, each of these has its own characteristics, but for details, please refer to the literature (various discussions on mouthpieces and greening methods), Materials of the Institute of Electronics and Communication Engineers, pRL75.
-66, etc.), so a detailed explanation thereof will be omitted here.

Finally, the collation determination unit 5 is composed of a block 50 with the same name and a fingerprint registration file 5■.

Here, in order to speed up the processing, the matching determination unit 50 performs matching by dividing it into rough matching and fine matching, as shown in FIG.

First, as shown in Fig. 9(a), a template for rough matching that covers a small area is used to count the number of intersections, and this is applied over the entire candidate area in the test fingerprint image. This is performed by sequentially shifting the positions one pixel at a time, and calculating the number of intersections at each position, that is, the matching error. Then, the parts (plurality) obtained in this way where the matching error is below a certain value are set as the positions for fine matching, and as shown in FIG. Perform detailed matching using a matching template.

In this way, by first performing rough verification before fine verification, it is possible to narrow down the locations for fine verification, and therefore, according to this embodiment, the amount of data processing can be saved. In addition, since the template for rough matching itself is small, the amount of data processing is small, resulting in efficient processing as a whole.

Next, the matching and determining process by the matching and determining section 50 will be explained in more detail with reference to the flowchart in FIG.

Although not shown in this flowchart, a template is first created and stored in the fingerprint registration file 51 with a fingerprint designation number attached thereto.

The template at this time is, for example, by converting the valleys of the registered fingerprint into core wires, and if the result of core wire creation is 8 connections as shown in FIG. 11(a), the core wires are as shown in FIG. 11(b). A process is performed to enlarge the vicinity in three directions and leave the AND result, thereby converting it into a 4-connection as shown in FIG. 3(C) and storing it as a template. This is to prevent the intersection of 8-connected core wires from being unable to be detected by a simple AND operation, as shown in FIG. 12(a). In other words, as shown in FIG. 12(b), 8
This is because a concatenation and a 4-concatenation can be detected using an and.

In addition, in the above AND operation, the number of pixels that appear as overlapping core lines is first found, but in order to convert this to the number of intersection points, for example, an algorithm that counts the number of particles (for example, the number of red blood cells) in the image is used. An example of this is the "number recognition method" disclosed in Japanese Patent Application Laid-Open No. 51-41578. However, the number of overlapping pixels is approximately proportional to the number of steps, so instead of finding the number of intersections, the easily determined number of overlapping pixels may be used.

Returning to the flowchart in FIG. 10, when this process is started, first, the ridges of the input inspection fingerprint image are converted into core lines and read into the verification determination section (Sl, S represents step), and then Then, a rough verification template is read according to the fingerprint designation number (S2).

Note that the small area for rough matching at this time is, for example, the first
It is sufficient to use the size shown in FIG. 3, but this is just one example, and it goes without saying that the size, shape, etc. may be adjusted as appropriate to improve the processing speed.

Next, the template and the cored inspection fingerprint image and
, and the number of pixels that become II I II is counted in the rough matching area and set as the number of intersection pixels CN (S3).

If the number of intersection pixels exceeds the fixed value TH, it is assumed that the matching at that position did not fail (S4), and the matching position is shifted to the next position (by one pixel), (S1L, 512), the same process.

If all positions are tested as matching positions and only mismatches are detected, the test fingerprint is determined to be different from the registered fingerprint.

An example of how the template moves (route) for rough matching at this time is shown in FIGS. 14(a) and 14(b).

If the number of intersections is equal to or less than the constant value THI in the rough matching (result in S4 is yes), it is assumed that the rough matching is a match, and the process proceeds to the fine matching to find the size of the matching area.

In fine matching (S5 to S8), compared to rough matching, the size of the matching area is sequentially enlarged to calculate the number of intersection pixels, and when this value exceeds a certain value, T Ht, the size of the matching area is calculated. , the size of the matching area at the matching position (39,5
10).

As for the overall judgment process, the change in the size of the matching area when the matching position is shifted one pixel at a time is investigated, and the largest change is taken as the size of the matching area between the test fingerprint and the registered fingerprint. .

FIG. 15 shows an example of enlargement of the verification area.

In FIG. 15, it can be seen that the central area is the area used in rough matching, and in fine matching, the matching area is gradually expanded around it.

In this way, the size of the matching area is determined through the above processing, but if only rough matching is completed at this time, the fingerprints do not match. On the other hand, if the size of the matching area is larger than a certain level, it can be determined that the fingerprints match, but the size of the matching area at this time is determined by a threshold value depending on the matching accuracy required for the judgment. All you have to do is decide.

〔Effect of the invention〕

As explained above, according to the present invention, since it is possible to perform fingerprint verification with sufficient accuracy through pattern matching, the problems of the prior art can be fully addressed and fingerprint verification can be performed with a simple configuration. This makes it possible to easily apply it to identity verification, etc.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the fingerprint verification method according to the present invention, FIG. 2 is an explanatory diagram of image data thinning, FIG. 3 is an explanatory diagram of the allowable width by thinning, and FIG. 4 is a diagram of pattern mating. Fig. 5 is an explanatory diagram of pattern matching judgment, Fig. 6, Fig. 7, and Fig. 8 are explanatory diagrams of the optical system, Fig. 9 is an explanatory diagram of the template, and Fig. 10 is an illustration of the book. Flowchart for explaining operation in one embodiment of the invention, 1st
Fig. 1 is an explanatory diagram of data conversion, Fig. 12 is an explanatory diagram of intersection detection, Fig. 13 is an explanatory diagram of a small area for rough matching, Fig. 14 is an explanatory diagram of the movement path of a rough matching template, and Fig. 15 16 is a block diagram showing a conventional example of a fingerprint matching system, FIG. 17 is an explanatory diagram of feature point extraction, and FIG. 18 is a flowchart for explaining the matching process.
Fig. 19 is an explanatory diagram of the minutiae, Fig. 20 is an explanatory diagram of the difference in the position of the minutiae, Fig. 21 is an explanatory diagram of the fingerprint reading device, and Fig. 22 is an explanatory diagram of the fingerprint reading device.
The figure is an explanatory diagram of matching errors in pattern matching. 1... Fingerprint reading unit, 2... Signal processing unit, 3... Intake determination unit, 4... Core wire extraction storage unit, 5... ...Verification and determination section, 10...
Light source, 11... Diffusion plate, 12... Prism, 13... Lens, 14... Video camera. Figure 2 (a) (b) Heart of Tanibe Figure 3 Heart of Ryukube To the Red Rat Ward (('('('('-) Total °Cl@a! Book 5 Q-1X4 Country No. 6 Figure 7 Figure 8 (0) (b) Figure 1I (Q) (b) (C
)(0)
(b) Fig. 13 Fig. 14 Fig. 15 Fig. 18 RLI 114F 罎gin combination Fig. 19 Tor: File fI 耘Φη A and Manji Fig. 20 Fig. 21

Claims (6)

[Claims] (1) In a fingerprint matching method in which registered fingerprint image data representing a fingerprint that has been registered in advance is compared with inspection fingerprint image data input by imaging the fingerprint to be inspected to determine whether the two match, one of the image data is Image processing means is provided to extract the ridges of the one and the troughs of the other as core lines, and a match between the two fingerprints is determined based on the number of intersections that appear between the two core lines by superimposing the two image data. A fingerprint verification method characterized by being configured as follows. (2) The fingerprint according to claim 1, wherein the match determination is performed based on the size of an area where the distribution density of the intersection points on the image plane is less than or equal to a predetermined value. Matching method. (3) The fingerprint verification method according to claim 1, characterized in that the number of intersections is determined by the number of pixels that appear overlapping each other in the core wires. (4) The fingerprint verification method according to claim 1 or 3, characterized in that the core wires are extracted as data in a four-connected configuration. (5) The fingerprint verification according to claim 1 or 3, wherein one of the core wires is extracted as data of a 4-connection configuration, and the other is extracted as data of an 8-connection configuration, respectively. Method. (6) In claim 1, each image data is captured when the number of black pixels within a predetermined range of captured data falls within a predetermined value. A fingerprint verification method characterized by: