JP2788529B2 - Dictionary registration method for fingerprint matching device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] The present invention relates to a dictionary registration method in a case where feature points of a fingerprint to be a dictionary are registered in a fingerprint collation device, and aims to improve the accuracy of the registered dictionary. The input fingerprint image is subjected to preprocessing and thinning processing, and a predetermined number of feature points are read from the processed fingerprint image. The pattern matching is performed by scanning a certain area of the above, the number of scanning positions at which the degree of mismatch is smaller than a predetermined threshold is counted, and only the feature points corresponding to the window whose count value is equal to or less than the certain value are obtained. The dictionary is registered as a true feature point.

[Industrial application field] The present invention relates to a dictionary registration method for registering a feature point of a fingerprint serving as a dictionary in a fingerprint collation device.

In recent years, with the introduction of computers into a wide range of social systems, stakeholders have been interested in system security. ID cards and passwords, which have been used as a means to identify themselves when entering a computer room or when using a terminal, have raised many questions in terms of security. On the other hand, fingerprints are considered to be the most powerful means of personal identification because they have two major characteristics, that is, everyone's identity and lifelong invariance, and much research and development has been conducted on a simple personal identification system using fingerprints. I have.

[Prior Art] FIG. 3 is a conceptual diagram showing the configuration of a conventional fingerprint collation apparatus. First, the operation at the time of registration will be described. A fingerprint pattern is detected by pressing a finger on the fingerprint sensor 1 and converted into digital data by an A / D converter (not shown) in the fingerprint sensor 1. The converted digital data (fingerprint data) is converted into binary data “0” and “1” by the following binary or circuit 2 and stored in the frame memory 3.

After the fingerprint data stored in the frame memory 3 is sequentially read out, the fingerprint data enters the feature information extraction / collation circuit 4 where the feature information is extracted. Here, the characteristic information means, for example, a branch point as shown in FIG. 4 (a), an end point as shown in (b), and the like. A fingerprint can be specified by determining where and how many such branch points and end leads are located. The feature information extraction / collation circuit 4 performs preprocessing and thinning processing on the data read from the frame memory 3. Here, the pre-processing refers to removal of noise and sweat glands included in the fingerprint image. Further, the thinning processing refers to making a ridge or the like of a fingerprint image a thin line so that the pattern can be easily viewed. Then, feature information is extracted from the fingerprint image that has been subjected to the preprocessing and the thinning processing.
The extracted characteristic information is stored in the personal characteristic information data file 5
Is stored in The above operation is repeated for a plurality of individuals, and individual characteristic information is stored in the individual characteristic information data file 5.

When the registration of the characteristic information amount in the personal characteristic information data file 5 is completed in this manner, the operation for collating the fingerprint of the individual is started. In the case of collation, when a finger (a finger previously used for registration, for example, an index finger) is pressed against the fingerprint sensor 1, a fingerprint pattern is detected in the same manner as at the time of registration, and an A / D converter (shown in FIG. ) To convert to digital data. The converted digital data is converted into binary data “0” and “1” by the following binarization circuit 2 and stored in the frame memory 3.

The feature information extraction / matching circuit 4 reads the fingerprint data for matching stored in the frame memory 3 and the feature information for each individual stored in the personal feature information data file 5, and performs matching (pattern matching) of both. Do. When the number of matches of the feature information is equal to or more than a predetermined number, it is determined that the fingerprints match. Here, the feature information extraction operation of the feature information extraction / collation circuit 4 will be considered. The feature information extraction / collation circuit 4
First, pre-processing (processing such as noise removal and sweat gland removal) and thinning processing are performed on the input binary data, and characteristic information is extracted from a fingerprint image that has been subjected to such processing.

[Problems to be Solved by the Invention] By the way, if there are many noises and sweat glands in the preprocessing stage, these noises and sweat glands are read as pseudo feature points. Since such a pseudo feature point is not an original feature point of the fingerprint, if such a pseudo feature point is registered as a dictionary,
The accuracy of fingerprint verification is lost.

FIG. 5 shows true feature points, and FIGS. 6 (a) to (d) show pseudo feature points. In FIG. 6, (a) shows a bridge, (b) shows a beard, (c) shows a sweat gland, and (d) shows a crack.

As is clear from a comparison between the true feature points shown in FIG. 5 and the pseudo feature points shown in FIG. 6, the true feature points disturb the periodicity of the surroundings. On the other hand, the pseudo feature points do not disturb the periodicity of the surrounding images. Therefore, FIG.
Consider windows of feature points or pseudo feature points as shown in the figure, scan the area around the feature points with these windows, and perform pattern matching. In the case of true feature points, FIG. While the characteristic of the degree of inconsistency as shown is obtained, in the case of a pseudo feature point, the characteristic of inconsistency as shown in FIG. 8 is obtained.

As is clear from FIG. 7, in the case of a true feature point, only one portion where the degree of mismatch is close to 0 occurs. On the other hand, in the case of the pseudo feature point shown in FIG. 8, a portion where the degree of mismatch is small periodically occurs, and a portion which matches the feature point of the window has such a characteristic that the degree of mismatch is close to zero. . A broken line in the figure indicates a threshold.

The present invention has been made in view of such a problem, and an object of the present invention is to provide a dictionary registration method of a fingerprint matching device that can improve the accuracy of a registered dictionary.

[Means for Solving the Problems] FIG. 1 is a flowchart showing the principle of the method of the present invention. According to the present invention, a fingerprint image is input (step 1), preprocessing and thinning processing are performed on the input fingerprint image (step 2), and a predetermined number of feature points are determined from the processed fingerprint image. Only (Step 3), pattern matching is performed by scanning a certain area around the read feature point as a window (Step 4), and counting the number of scanning positions where the degree of mismatch is smaller than a predetermined threshold value Only a feature point corresponding to a window whose count value is equal to or less than a certain value is registered in the dictionary as a true feature point (step 5).

[Operation] A predetermined area around the read feature point is scanned using the read feature point as a window, pattern matching is performed, and the number of scanning positions at which the degree of mismatch is smaller than a predetermined threshold is counted. If the count value is equal to or less than a certain value, it is determined that the feature point is a true feature point and registered as a dictionary. In other cases, the dictionary registration is rejected as a pseudo feature point. Thus, according to the present invention, the accuracy of the dictionary can be improved.

[Example] Hereinafter, an example of the present invention will be described in detail with reference to the drawings.

FIG. 2 is a diagram showing an example of a system configuration for implementing the present invention, which shows a fingerprint collation device. In the figure, reference numeral 11 denotes a fingerprint sensor for reading a fingerprint image, for example, a CCD. 12
Is a numeric keypad for inputting commands and the like at the time of fingerprint reading, and 12 is a feature extraction / matching unit that extracts feature information from a fingerprint image read by the fingerprint sensor 11 and matches the read fingerprint with a dictionary. It is.

14 is a feature information storage unit that stores the feature information extracted by the feature extraction / collation unit 13, and 15 is a pattern matching unit that scans a certain area around the feature point window using a feature point window when storing the feature information. A dictionary evaluation unit that counts the number of scanning positions where the degree of inconsistency is smaller than a predetermined threshold value, and registers only feature points corresponding to windows whose count value is equal to or less than a predetermined value as true feature points in the dictionary. This is the part that characterizes the present invention. To explain the operation of the circuit configured as described above,
It is as follows.

(Dictionary registration) First, the finger to be registered (for example, index finger) is
When put on the fingerprint sensor 11, the fingerprint sensor 11 reads the fingerprint image of the finger and converts the fingerprint image into an electric signal (analog signal). The fingerprint image signal converted into the electric signal enters the feature extraction / collation unit 13. Here, when a fingerprint registration command is entered from the numeric keypad 12,
The feature extraction / collation unit 13 starts processing of the fingerprint image read from the fingerprint sensor 11. First, the fingerprint image read by the feature extraction / collation unit 13 is converted into digital data by an A / D converter (not shown). Next, the fingerprint image converted into digital data is converted into binary data of “0” and “1” using a predetermined threshold. The converted binary data is stored in a built-in frame memory (not shown).

Next, the feature extraction / collation unit 13 performs preprocessing and thinning processing on the binary data read from the frame memory.
Here, the pre-processing refers to processing such as noise removal and sweat gland removal as described above. The thinning process refers to making a ridge thin to clarify a pattern. When such processing is completed, the feature extraction / collation unit 13 extracts feature information from the fingerprint image that has been processed. The extracted feature information is sequentially sent to the dictionary evaluation unit 15.

The dictionary evaluation unit 15 performs pattern matching by scanning a predetermined area around the input feature point window, and counts the number of scanning positions where the degree of inconsistency is smaller than a predetermined threshold. Then, only the feature points corresponding to the window whose count value has become equal to or less than the certain value are registered in the dictionary as true feature points. As described with reference to FIGS. 7 and 8, in the case of a true feature point, a certain area around the feature point is scanned by the window of the feature point, and when the pattern matching is performed, the degree of mismatch is determined. When the number of scanning positions where is smaller than a predetermined threshold value is counted, the counted value becomes equal to or less than a certain value.

On the other hand, in the case of a pseudo feature point, a certain area around the feature point is scanned in the window of the feature point, and when pattern matching is performed, the degree of inconsistency is smaller than a predetermined threshold. When the number of positions is counted, the counted value becomes larger than a fixed value. Therefore, the dictionary evaluation unit 15 performs evaluation by checking the degree of inconsistency of the feature points and determines whether to register as a dictionary. In the case of a feature point to be registered, it is registered as a dictionary in the feature extraction / collation unit 13 information storage unit 14. Thus, according to the present invention, the characteristic information storage unit 14 can always store probable characteristic information.

(At the time of collation) At the time of collation, a finger is placed on the fingerprint sensor 11, and then the user enters his / her ID number using the numeric keypad 12. As a result, the feature extraction / collation unit 13 determines the search range of the feature information in the feature information storage unit 14 based on the ID number and reads it out at the time of comparison, thereby narrowing the search range.

The fingerprint image read by the fingerprint sensor 11 is binarized by the feature extraction / collation unit 13. The binarized data and the feature information (dictionary) stored in the feature information storage unit 14 are subjected to pattern matching (matching) by the feature extraction / matching unit 13. And
If the number of feature points subjected to pattern matching is a predetermined number, the feature extraction / collation unit 13 determines that the fingerprints match.

[Effects of the Invention] As described above in detail, according to the present invention, pattern matching is performed by scanning a certain area around the extracted feature point window, and the degree of mismatch becomes smaller than a predetermined threshold value. The number of scanning positions is counted. If the count value is equal to or less than a certain value, it is determined that the feature point is a true feature point and registered as a dictionary.
In other cases, the dictionary registration is rejected as a pseudo feature point. Thus, according to the present invention, the accuracy of the dictionary can be improved.

[Brief description of the drawings]

 FIG. 1 is a flow chart showing the principle of the method of the present invention, FIG. 2 is a diagram showing an example of a system configuration for implementing the present invention, FIG. FIG. 5 is a view showing an example of information, FIG. 5 is a view showing a true feature point, FIG. 6 is a view showing a pseudo feature point, FIG. It is a figure which shows the mismatch degree characteristic of a feature point. In FIG. 2, 11 is a fingerprint sensor, 12 is a numeric keypad, 13 is a feature extraction / collation unit, 14 is a feature information storage unit, and 15 is a dictionary evaluation unit.

────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Taku Niizaki 1015 Uedanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Prefecture Within Fujitsu Limited (58) Field surveyed (Int.Cl. ⁶ , DB name) G06T 7/00

Claims (1)

(57) [Claims] 1. A fingerprint image is input (step 1), a pre-processing and a thinning process are performed on the input fingerprint image (step 2), and a predetermined number of feature points are determined from the processed fingerprint image. A number of readings (step 3), pattern matching is performed by scanning a certain area around the image using the image near the read feature point as a window (step 4), and a scanning position at which the degree of inconsistency is smaller than a predetermined threshold value Wherein the feature point corresponding to the window whose count value is equal to or less than a certain value is registered in the dictionary as a true feature point (step 5). Registration method.