JP2575676B2 - Automatic Matching Point Tracking Method for Fingerprint Matching - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a fingerprint collation method for personal identification, for example, as a finger key (key by fingerprint collation) instead of a door opening / closing or an IC card / cash card. What is used.

(Summary of the Invention) In the present invention, in a method for identifying the identity of two fingerprint images whose imaging positions are shifted, a collation start reference point (for example, an approximate center point) of both images is determined, and the reference point is set as a center. 3 or more search ranges are provided concentrically, and each range is 4 to 8
Divide into blocks.

The inside of the innermost circle is defined as a defined area and divided into four fan-shaped blocks. The middle circle is defined as the first inference region, and 4 to 8
Divide into two sector blocks. The inside of the outermost circle is divided into eight fan-shaped blocks, and is set as a second inference region. The matching point search is started from the above-mentioned reference point, and is successively moved until the reference point is moved to the first inference area and the second inference area, two images are compared, and it is determined that the user is identified. However,
If there is no point showing a matching rate equal to or more than a certain value in both the specified area and the first inference area, the two images are determined to be different fingerprints, and the search is stopped.

The movement of the search range from the specified region to the first inference region and from the first inference region to the second inference region is as follows. In this case, the process is performed on the block outside the block.

If it is predicted that there is a matching point further outside the second inference area, a similar search is restarted using the highest matching rate point as a new search starting point.

(Prior Art) Conventionally, personal identification numbers and seals have been used as means for personal identification, but the former can be used by others if the personal identification number is leaked, and the seal is used by others. In addition, it is possible with a recent printing technique to create a seal image copy of a seal itself, which is almost indistinguishable from a genuine seal.

The fingerprint is unique to the individual, and it is difficult to make an image copy. Further, in the case of a fingerprint, as described in another application, the image copy can be distinguished from the real copy to some extent. There is no such means for identification.

There is already a fingerprint identification device for criminal investigation, but this device is for identifying a specific person from an unspecified number of people, the purpose of which is to select a third party, and the method of image recognition is It is based on fingerprint feature point extraction and matching (minutiae network method).

Also, fingerprints collected from the field are often incomplete,
In order to pick up limited fingerprint information without giving it a small size, it is a large-scale system using a large computer.

When the user himself / herself tries to use a fingerprint as a key as in the present invention, almost complete fingerprint information is obtained.
However, in order to search for a matching coincidence point using the same amount of information by the full-scale matching method, for example, a minute unit time is required using a small CPU, which is not yet practical.

On the other hand, attempts have been made to extract the characteristic points of the fingerprint instead of full-scale collation, and to try to identify the individual from the pattern of the positional relationship between the characteristic points.However, the reproducibility of the characteristic points largely depends on the finger and the state of the impression. Therefore, there are many factors that have not been collated yet, and the process of extracting feature points is complicated and takes a long time.

(Problems to be Solved by the Invention) The present invention provides an inexpensive and practical fingerprint identification method for personal identification that can be processed in a practically short time (within 2 to 3 seconds) and with a small CPU. It is an object.

The present invention further focuses on the fact that the same fingerprint is scarcely present when the fingerprint is cut out from the representative pattern portion by an effective size and compared, and the processing time is shortened by performing full-scale matching of the cut portion. It is still another object of the present invention to provide an efficient method for automatically tracking a matching point for finding a matching point between two fingerprint images in the shortest time, that is, with the minimum number of processings.

(Means for Solving the Object) The fingerprint image used in the present invention is optically imaged in a size of about 13 mm × 10 mm, and the image is a solid-state image sensor (CCD)
Are stored as binarized outputs. The binarized output is equivalent to 0.05 mm / bit. The image actually used for collation is about 5m representing the representative pattern part of the fingerprint in the image taken above.
a verification image of m ^2. The extraction of the representative portion is performed by reading the ridge density (the number of ridges / unit) of the captured image as described in another application, and cutting out the average address of a region where the density exceeds a certain value as a center point.

In order to identify the identity between the previously registered master image and the newly input sample image, the above-mentioned matching images are entirely matched. The problem is that since the master image and the sample image are usually misaligned, how to find the matching coincidence point quickly and efficiently at the shortest distance.

(Example) As shown in FIG. 1, the number of ridges / unit is read for each of the XY axes and the change is analyzed to obtain a constant density distribution for each of the two axes of the captured fingerprint image (256 × 192 bits) as shown in FIG. The average address over the range is determined, and this is defined as the approximate center point. The approximate center point is an approximate value that fluctuates due to an imaging region, a change in image quality (missing of an image edge), or the like. This approximate center point is used for the first alignment when actually comparing two fingerprint images. The master image and the sample image are first fully matched with respect to both matching images (96 × 96 bits) cut out from the approximate center point (all values of the binary output of the image are compared and matched). If you are lucky, this first match will identify you. Normally, a satisfactory collation rate cannot be obtained at the approximate center point, and the collation is repeated by shifting the registration reference point of the master image little by little around the central point, and a collation coincidence point is searched for.

Conventionally, for example, a huge amount of processing time was required because the matching process was sequentially performed until the correct answer was obtained by shifting one bit at a time from the end of the image. However, in the case of the present invention, the addresses of both images are almost the same. Since the matching search can be started from the matching approximate center point, the time until the matching matching point is reached is short. In addition, since it is experimentally confirmed in which range from the approximate center point the matching coincidence point is located, it is easy to judge that it is another fingerprint simply by searching only within a certain range. is there.

It has been confirmed that when a large number of two images are collated, the distribution of the collation rate is as shown in FIG. That is, in the case of a personal fingerprint, the number of occurrences of the collation rate of 70-96% is 95
Account for%. On the other hand, in the case of other-person fingerprints, the number of occurrences in which the matching rate exceeds 70% is less than 0.1% of the total number. From this result, if the collation rate of 80% is used as a practical reference value (the criteria for judging the collation rate of 80% or more as a person), it is determined that the person is a sufficient person without misidentifying a fingerprint of another person.

As described above, in the present invention, the fingerprint image is first subjected to the collation processing using the approximate center point as a cross reference. Normally, the approximate center point and the matching coincidence point of the two images deviate from each other.

Therefore, a certain range (radius
A search for a matching point is performed while shifting the matching reference point of the master image to within a 10-bit circle). FIG. 3 A _{1 to} D ₁
Is defined as a defined area, and the entire matching of the matching image is continued in a defined order for each point in a circle having a radius of 10 bits. If there is a point indicating a collation rate of 80% or more during this search, it is determined that the user is the principal and the collation processing ends.

When there is no point showing a matching rate of 80% or more in the specified area, the specified area is divided into four blocks (A _{1 to} D _{1 in the first to} fourth quadrants) as shown in FIG. The distribution of search points (for example, points of 60% or more) where the matching ratio exceeds a certain value is accumulated, and this search point ratio is defined as a prediction coefficient. In addition,
It has been experimentally confirmed that there is a high probability that the matching point of the two images exists in the radical direction of the region where the prediction coefficient is high. Therefore, the outer circle of the block indicating the highest predictive coefficients (of A ₂ to D ₂ of FIG. 2, for example, for high prediction coefficient block A ₁ A ₂₎ is then searched for matching coincident point Can be predicted as the most effective area for performing In the second search area, a circle with a radius of 19 bits (A _{2 to} D _{2 in} FIG. 3)
Is defined as the inference area. As before, when there is no point showing a matching rate of 80% or more in the search block (A _{2 in} FIG. 3) of the first inference area, this block is divided into two (A _{21 in} FIG. 3 and A 2 in FIG. 3). A ₂₂ ), the search is prioritized out of the two blocks with the higher prediction coefficient. This is a second inference area having a radius of 25 bits indicated by A ₃₁ , A ₃₁ ... D ₃₁ and D ₃₂ in FIG.

For example, it advances the search area from A ₁ to A _2, further when proceeding to A _31, the prediction coefficients in the A ₃₁ region is lowered than prediction coefficients A ₂₁ regions, and the A ₂₂ region prediction coefficient is generated May not. In this case, the inference is returned to the prescribed area, and the first inference area (in this case, D ₂ ) located outside the block (in this case, D ₁ ) having the next highest prediction coefficient after A ₁ , And a second inference domain (D _{32 in} this case)
To the search area. This is the process of automatic tracking that often occurs when a matching match point exists near the left side (the boundary area between A and D) on the Y axis experimentally.

As a result, it did not exist matching ratio of 80% or more points D ₃₂ region, when the prediction coefficients becomes higher, collation coincidence point is expected to further present outside the D ₄ region.
This rather than expanding the search area in the D ₄ region if, as the last in the search areas (in this case D ₃₂ region) most matching a new starting point higher was the point of rate in the same as the search to this Do it again. This is called automatic update tracking.
This is more to expand the search area to the D ₄ area, because there is better to re-start new move search start point is efficient search.

When there is no point indicating the matching rate of 60% or more in the specified area, the search is performed by moving the matching reference point to all the blocks in the first inference area (the first inference area in this case is the specified area). Treated as an extension). This first inference area also has 60
When there is no point showing the matching rate of not less than%, the search is stopped, and both matched images are determined to be different fingerprints. Because the master image and the sample image use the approximate center point as the first mutual reference point, the positions of the two images already substantially match. Therefore, in the case of a personal fingerprint, at least a prediction coefficient should be generated in the specified area or the first inference area.

However, even when all the steps of the automatic matching point tracking according to the present invention have been completed, there is a case where there is no point showing a matching rate of 80% or more finally. In this case, the probability of identity is high as shown in FIG. 2, but this is an example in which images are not matched due to image distortion or the like caused by the degree of finger pressing when capturing a fingerprint image. In this case, as described in another application, correction and collation for correcting image distortion and performing collation are performed. It has been experimentally confirmed that if the secondary determination is performed by the correction collation, there is almost no case where it is impossible to determine whether the user is the person or the other person.

On the other hand, in the process of searching for a matching match point, there is a point where the matching rate is high even if the two images do not completely match. Generally, the matching rate increases as the matching point is approached. Even if the two screens do not completely match, if the collation rate exceeds 80%, there is no problem in determining that the user is the person. Naturally, the matching process is completed when the matching rate exceeds 80%.

(Effect of the Invention) The automatic matching point tracking (searching) method of the present invention has the following advantages.

(1) In the case of the present invention, the search is started by first performing the overall matching of the matching image by matching the approximate center points of the two-sided images. Therefore, in the case of a personal fingerprint, the two images have a positional relationship quite close to the matching point at the start of the matching, and the search efficiency is much higher than the conventional method of sequentially searching for the matching matching point from the end of the image.

(2) When the comparison of both images is performed using the approximate center point as the search start point, the determination in the case of another person's fingerprint can be performed only by searching within a certain range near the approximate center point, and the determination time is very short. Becomes shorter.

(3) The automatic matching point tracking method in the present invention
The prediction coefficient is calculated in parallel with the search for the matching point in the area, and the direction of the matching point can be immediately predicted to determine the next search area with the highest probability. The search efficiency is much better than the method of expanding the search area omnidirectionally.

(4) When the matching point is predicted to be outside the second inference area, the search time can be shortened by automatically updating the search reference point and restarting the search without unnecessarily expanding the search area.

[Brief description of the drawings]

FIG. 1 shows an embodiment for obtaining an approximate center point according to the present invention,
FIG. 2 shows an example of the matching rate distribution. FIG. 3 is a diagram for explaining an area of matching point tracking (search) used in the present invention. C: approximate center point, A _{1 to} D ₁ ... defined area, A _{2 to} D ₂ ... first inference area, A ₃₁ , A _{32 to} D ₃₁ , D ₃₂ ... second inference area.

Continued on the front page (72) Inventor Hiromi Irie 2-11-14, Suwa-cho, Higashimurayama-shi, Tokyo (72) Inventor Ichio Nakahata 5-7504 Takamori, Isehara-shi, Kanagawa Prefecture (72) Inventor Yasuyuki Asaba Kanagawa (72) Michio Sasaki Inventor 1736-3 Kanamori, Machida-shi, Tokyo Examiner Teruhisa Chiba

Claims (5)

(57) [Claims] In a method for comparing a master fingerprint image and a sample fingerprint image to identify their identity, a reference point (for example, an approximate center point) for starting the comparison is determined, and a concentric circle is formed around the reference point. At least three search ranges are set, a circle near the center point around the reference point is divided into four blocks of a fan-shaped area (defined area), and 4 to 8 blocks are formed in the circle excluding the defined area outside the circle. Fan-shaped area (first inference area)
And further divides the inside of the circle excluding the outside of the defined region and the first inference region into eight blocks of fan-shaped regions (second inference regions), and defines the defined region, the first inference region, and the second inference region. A matching point automatic tracking method characterized in that a matching reference point of a master image is moved in the order of regions and a search is performed while predicting a matching point. 2. The method according to claim 1, wherein when there is no matching coincidence point in the specified area, a first inference area outside the specified area having the highest prediction coefficient is searched next. Automatic tracking method of the collation point described in the item. 3. The method according to claim 1, wherein when there is no point indicating a matching rate equal to or higher than a predetermined reference in both the specified area and the first inference area, the two images are determined to be different. The method for automatically tracking a collation point according to claim 1. 4. A collation coincidence point does not exist in the first inference region. However, when a prediction coefficient is generated, the prediction coefficient is compared with the prediction coefficient in the specified region. 2. The method according to claim 1, wherein the second inference area outside the first inference area is preferentially searched, and when the number decreases, another block in the first inference area is searched next according to the prediction order in the specified area.
Automatic tracking method of the collation point described in the item. 5. A case where no matching coincidence point exists in both the specified area and the first and second inference areas and the prediction coefficient is sequentially increased (when a matching coincidence point exists outside the second inference area). 5. The collation according to claim 4, wherein the automatic tracking of the first to fourth items is restarted using the highest collation rate point in the final search area as a new collation start point. Point automatic tracking method