JPH02187866A - Method and device for collating individual - Google Patents

Abstract

PURPOSE:To improve a collation ratio by comparing individual feature information being stored with the shape of feature points seen from the front image of a human face. CONSTITUTION:A control part 5 reads out six pieces of individual feature information stored in an individual feature storing part 4 based on an individual code inputted in an individual code input part 1, inputs the pieces of read out information to a collating part 34, and calculates the similarity degree between the pieces of information. The pieces of calculated similarity degree information are inputted to a deciding part 35, the average similarity degree of the six similarity degrees is calculated, and the average similarity degree is made into the integrated similarity degree of face information. In the deciding part 35, the integrated similarity degree is compared with a decision value set and stored for each individual, and when the integrated similarity degree is larger than or the same the decision value, it is decided that a user is a registered principal.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to an individual verification method and device used for controlling access to a controlled area, etc., and in particular, relates to an individual verification method and device used for controlling access to and from a controlled area. In particular, the present invention relates to an individual verification method and apparatus that performs individual verification using image features such as the inner and outer corners of the eyes.

[Prior art and problems to be solved by the invention] Various attempts have been made to identify individuals, one of which is a simple one that uses an identification card, and one that uses fingerprints with the recent development of image signal processing technology. More recently, methods using images of faces have been proposed.

Methods that use facial images can be roughly divided into personal identification methods based on profiles and silhouettes, and personal identification methods based on normal frontal faces. July 1, 1988
5th, IEICE, PRII8g-46)
.

Personal identification methods using profile silhouettes have the problem of lacking stability in personal identification.

Personal identification methods based on frontal normality include methods that use features related to the position of facial parts and features related to shape.

Regarding such technology, Japanese Patent Application Laid-Open No. 63-118
No. 473 (“Image-capturing type unlocking/locking device”) is known. What is disclosed in Japanese Patent Application Laid-Open No. 118473/1983 extracts the shape and arrangement of human eyes as features, and determines whether they match feature data registered in advance. More specifically, the distance between the left and right eyes, the vertical and horizontal widths of the left and right eyes, or the ratio thereof is used. However, this technique uses parts that change depending on facial expressions, and there is still a problem in that verification stability is lacking.

In addition, since image processing uses a lot of data for boat fishing, the problem arises that when the number of objects to be identified increases, it takes a long time for verification.

An object of the present invention is to provide an individual verification method and apparatus that improve the accuracy of individual verification and shorten the verification time.

[Means and effects for solving the problem] The present invention basically provides a system that has large individual differences and remarkable characteristics, has a high degree of stability in verification, and requires fewer objects to be identified.
Among the L F (Local Feature) features,
Individual matching is achieved by comparing the similarity with pre-registered standards using the image data of the areas near the right inner corner, left inner corner, right outer corner, and left outer corner as feature data. .

According to the first aspect of the present invention, an image registration step of registering in advance image data of areas near each of at least the right inner corner of the eye, the left inner corner, the right outer corner, and the left outer corner of the individual to be matched; an image input step in which facial image data is generated by optically scanning the face of an individual; A matching part extraction step of extracting image data of areas near each corner of the eye, and comparing and matching the image data extracted in the matching part extraction step with the pre-registered image data to determine the degree of similarity between them. An individual matching method is provided, which includes a similarity calculation step of calculating the similarity, and a determination step of determining the individual if the similarity calculated in the similarity calculation step is greater than or equal to a predetermined value.

Each pixel of the image data in the image registration stage and the image input stage is multivalued image data, and the similarity is calculated for the multivalued image data.

The similarity calculation is performed by calculating the inner product of the vector representation of the registered image data and the vector representation of the part-extracted image data, and the length of the vector representation of the registered image data and the part-extracted image. This is done by dividing the data by the product of the length of the vector representation.

Between the image registration stage and the image input stage, a personal code input and identification stage in which a personal code for identifying the person to be verified is input, and the verification process is proceeded only when the input personal code is normal. The image input step, the matching part extraction step, the similarity calculation step, and the determination step are performed based on a personal code.

The image registration step further registers in advance, as LF features, image data of regions in the vicinity of the right eyebrow medial end point and the left shoulder medial end point, and the verification region extraction step further registers the image data obtained in the image input step. Image data in areas near the right eyebrow inner end point and left shoulder inner end point are also extracted from the image data, and similarity determination is also performed in the vicinity of the right eyebrow inner end point and left shoulder inner end point. This improves the accuracy of personal matching in the negative layer compared to the case where only the image data in the vicinity of the right inner corner, the left inner corner, the right outer corner, and the left outer corner are used.

In addition, as another aspect of the present invention, there is provided an apparatus for implementing the above-described method, that is, image data of at least the areas in the vicinity of the right inner corner of the eye, the left inner corner, the right outer corner, and the left outer corner of the individual to be matched. an image input means for optically scanning the face of the individual to be matched to generate facial image data; and an image input means for generating facial image data by optically scanning the face of the individual to be matched; A comparison part extraction means for extracting image data of areas near each of the left inner corner, right outer corner, and left eye corner, and comparing the image data extracted by the matching part extraction means with the pre-registered image data. Similarity calculation means for comparing and calculating the similarity between them;
A personal verification device is provided, which includes a determination device that determines the individual when the similarity calculated by the similarity calculation device is equal to or greater than a predetermined value.

Each pixel of image data in the image registration means and the image input means is multivalued image data, and the similarity is calculated for the multivalued image data.

The similarity calculation is performed by calculating the inner product of the vector representation of the registered image data and the vector representation of the part-extracted image data, and the length of the vector representation of the registered image data and the part-extracted image. This is done by dividing the data by the product of the length of the vector representation.

further comprising a personal code input and identification step of inputting a personal code for identifying the person to be matched and proceeding to the matching process only when the input personal code is normal, the image input process and the matching part extraction process; , the similarity calculation process and the determination process are performed based on the personal code.

The image registration means further registers in advance image data of regions in the vicinity of the inner end point of the right eyebrow and the inner end point of the left shoulder, and the matching region extraction means further registers the right eyebrow from the image data obtained by the image input means. The image data of the areas near the medial end point and the left shoulder medial end point are also extracted,
Similarity determination is also performed in the vicinity of the right eyebrow inner end point and the left shoulder inner end point.

〔Example〕

Preferred embodiments of the invention will now be described with reference to the accompanying drawings.

FIG. 1 shows a flowchart of a personal verification method according to an embodiment of the present invention. The personal verification method will be described below with reference to the same figure.

The personal verification method of the present invention can be broadly classified into registration processing of reference data for verification (step 10), and verification processing performed on the actual individual to be verified (steps 21 to 33).
We parted ways.

Step 10 This registration process is performed offline.

First, the face of the individual to be matched is photographed from the front in a normal state using a CCD camera, etc., and images are taken of the areas near the inner corners of the left and right eyes, the outer corners of the left and right eyes, and the inner end points of the left and right eyebrows. Extract the data.

As shown in Figure 2, this image data is extracted based on the individual to be matched (the same applies hereafter), left and right inner corner square areas C and A, left and right outer corner square areas and B, left and right eyebrows. This is done for the inner endpoint square regions E and F.

These square areas are, for example, m as shown in FIG.
=21Xn=21 pixels. However, in Figure 3,
As an example, a square region A at the inner corner of the right eye is shown.

Further, each pixel has 8 bits, that is, a gray scale (gray code) from 0 to 255. Therefore, image data photographed by a CCD camera is converted into digital values of 0 to 255 via an AD converter.

The image data obtained in this manner for the left and right inner corner square areas C and A, the left and right outer corner square areas and B, and the left and right eyebrow inner corner square areas F and E are saved as reference data for comparison. .

When creating reference image data for comparison, 5 obel
Create a contour image by applying 0perator.

In addition, a median filter is applied to absorb fluctuations in an individual's face in a normal state and remove random noise. Furthermore, the image data is normalized by performing rotation, enlargement, reduction, etc. using the center point between the inner corners of both eyes as the origin.

In order to improve the accuracy of the reference data for comparison, the above-mentioned photographing to creation of reference data for comparison is repeated several times, and the average value thereof is used.

When there are multiple individuals to be verified, the above registration process is performed for multiple individuals.

The verification process (steps 21 to 33) will be described below. This process is performed online each time, but basically the same process as the registration process described above is performed.

Step 21 First, the individual to be verified enters his or her personal code. Check whether the personal code is registered in advance. If the personal code has been successfully registered, the process moves to the following verification process; otherwise, the verification process is aborted.

This personal code input and identification process facilitates verification,
That is, this is intended to shorten the verification time and improve the confidentiality of personal verification.

In other words, by inputting a personal code and identifying that individual, misuse of individuals who are not subject to verification will be eliminated, and if a normal personal code is entered, reference data for verification will be identified. Therefore, the verification time can be shortened. Therefore, this personal code input is not essential to the verification method of the present invention itself. However, in the following explanation, steps 22 to 25 will be described based on the case where a personal code is used.If the personal code input is normal, the person's face with a normal facial expression, for example, can be captured using a CCD camera. to photograph.

A face image photographed by a CCD camera is converted to digital image data in which each pixel has a gradation of 0 to 255 via an AD converter. For this image data,
Furthermore, as mentioned above in the registration process, 5obel
0perator (operator) is applied to create a contour image.

Furthermore, in order to absorb the fluctuations in the individual's normal face and remove random noise, a median filter is applied to perform smoothing processing.

Furthermore, the image data is normalized by performing rotation, enlargement, reduction, etc. using the center point between the inner corners of both eyes as the origin. The purpose of normalization is to perform subsequent signal processing accurately and generally.

Steps 26 to 27 Next, a rectangular region of the facial parts is cut out and a feature extraction region is determined.

That is, as shown in FIG. 4, the contours are processed by the preprocessing unit 32 using the shape information of the eyes and eyebrows that have been set and input in advance, such as shape information such as being elongated horizontally, having left-right symmetry, and having continuity. Based on the face image information, both eyes and both layers are sequentially identified, and four rectangular areas containing one eye and one eyebrow are extracted (see FIG. 4).

Next, refer to the information on the rectangular area above and select the inner and outer corners of the eyes.
AA in Figure 4 to include the feature points of the shape of the eyebrow tip.
, BB, CC, DD, EE, and FF, a feature extraction region whose center is above the boundaries of the rectangular regions is set. Here, the shape of the inner and outer corners of the eyes, the outer corners of both eyes, and the shape of the two-layered head were used as feature points as a result of research by the present inventors. Compared to its characteristics, it is stable with little variation due to individual imaging conditions,
This is also because we have found that it expresses individuality well.

As described above, the left and right inner corner rectangular areas CC and AA correspond to each of the reference data for comparison.

Image data is extracted for the left and right eye corner rectangular areas DD and BB, and the left and right eyebrow inner end point rectangular areas FF and EE.

These left and right inner corner rectangular areas CC and AA, left and right outer corner rectangular areas DD and BB, and left and right inner corner rectangular areas FF and EE of the eyebrows are larger than the square area shown in FIG. 2, as shown in FIG. 4. Examples of this are shown in FIGS. 5(a) to (C), and the size thereof is, for example, β=60 X j =4
0=2400 pixels.

In this way, the rectangular area, e.g. AA, is larger than the square area for matching, e.g. A, because matching is performed on multiple individuals, so there are individual differences in the placement and size of eyes and eyebrows. This is to prevent discrepancies in verification by setting aside a wide range of data for verification even when looking at specific individuals.

Steps 28 to 30 Images of the left and right inner corner square areas C and A, the left and right outer corner square areas and B, and the inner corner square areas F and E of the left and right eyebrows, which are registered in advance in step 10 and correspond to the personal code. Data is read.

These pre-registered left and right inner corner square areas C and A,
Using each of the image data for the left and right inner corner square areas and B, and the left and right eyebrow inner corner square areas F and E, the left and right inner corner rectangular areas CC and AA, the left and right outer corner rectangular areas DD and BB, and the left and right The image data of the inner end point rectangular areas FF and EE of the eyebrows are compared and verified, and the degree of similarity is calculated for each.

This comparison and verification will be described with reference to FIGS. 5(a) to 5(C), using the right inner corner rectangular area AA and the right inner inner corner square area A as examples.

As described above, the right inner corner rectangular area AA is wider than the right inner inner corner square area A. Therefore, as shown in FIG. 5(a), first, the right inner corner square area A is made to coincide with the corner of the right inner inner corner rectangular area AA on a two-dimensional image plane, and in this state, for each pixel, 0 to Regarding 255 multivalued image data,
Calculate the similarity p. The algorithm for calculating the similarity is the following formula. Note that X or Y below is called the LP feature vector, and the LF feature, for example, This is a one-dimensional vector representation of a region obtained by rask scanning.

However, X-Y is the inner product of the vector representation of the image data for the right inner corner square area and the vector representation of the image data for the right inner inner corner rectangular area AA, and II X II・II Y 11 is the right inner corner square area A.
The lengths of the vectors of the image data of the right inner corner rectangular area AA expressed as vectors are shown.

As shown in Figure 5(b), the above similarity calculation is performed by shifting I pixels in the horizontal direction, and once the calculation is completed in the horizontal direction, it is shifted by 1 pixel in the vertical direction, and then shifted by 1 pixel in the Chinese direction. and do it.

This position shifting and matching operation is performed until the right corner is reached, as shown in FIG. 5(C).

In this way, the degree of similarity is calculated for the image data of the right inner corner square area A and the right inner inner corner rectangular area AA. In this case, the right inner corner square area A and the right inner inner corner rectangular area AA
The degree of similarity for the entire image data is determined by taking the average of the entire position-shift matching operation. Then, the maximum similarity among the calculated similarities is used as the final similarity. Therefore, FIGS. 5(a) to (
In the state shown in C), if the calculation is in progress and the similarity is extremely low, it is clearly not the target, so if you want to shorten the matching time, you can stop the calculation midway and proceed to the next step.

Similarities are calculated for other areas as well.

Then, the overall similarity of all these is determined by taking the overall average.

Steps 31 to 33 Compare the overall similarity and a predetermined judgment value. For example, if the determination value is 0.98, and the total similarity is 0.98 or more, it is determined (identified) that the person is the real person (step 33); otherwise, the person is not recognized as the real person (step 32).

In the above embodiment, as a preferred embodiment, similarity calculation is performed on the left and right inner corner rectangular regions CC and AA, the left and right outer corner rectangular regions DD and BB, and the inner end point rectangular region F of the left and right eyebrows.
Although the case has been described for the image data of F and EE, in implementing the present invention, basically, the minimum LF feature points excluding the inner end point rectangular areas FF and EE of the left and right eyebrows, that is, the left and right It has been found that the person can be identified with considerable accuracy even when the inner corner rectangular areas CC and AA and the left and right outer corner rectangular areas DD and BB are examined. In this case, the matching time is shortened.

Note that in the present invention, as described above, it is only necessary to calculate the degree of similarity for a maximum of six locations, so the matching time is short.

Next, a verification device that implements the above-described verification method will be described.

FIG. 6 is a block diagram of a personal verification device according to the present invention.

Reference numeral 1 denotes a personal code input section comprised of a numeric keypad or the like for the operator to input his or her own unique personal code. Reference numeral 2 denotes a face image input unit configured with a CCD camera or the like for capturing and inputting an image of the front face of an individual. Reference numeral 4 denotes a personal feature storage unit that stores facial feature information of an individual, which has been imaged and registered in advance, together with a personal code.

3, the face image inputted by the face image input means 2 is compared with the personal characteristic information inputted and stored in the personal characteristic storage section 4 in advance, that is, the reference data for the above-mentioned up to six types, to check whether the person is the real person or not. An interface 31 which indicates a matching means, performs A/D conversion of the face image data inputted by the face image input unit 2 into digital data, and interfaces each means, and processes the A/D converted face image data at the interface 31. A pre-processing unit 32 extracts it as a contour image, further removes random noise, and normalizes it.The pre-processing unit 32 extracts it as a contour image, and extracts it from the contour face image processed by the pre-processing unit 32. Cut out four rectangular regions in which , and based on the rectangular regions, extract six feature extraction regions including six feature points representing the inner and outer corners of both eyes, the outer corners of the eyes, and the inner corners of the eyebrows, that is,
Area extraction unit 33 that sets rectangular areas AA-FF in FIG.
Then, the six characteristic information of the individual corresponding to the above-mentioned six feature extraction regions, that is, the image data of the square areas A to F in FIG.
A matching unit 34 calculates the degree of similarity by matching the six feature extraction regions extracted in , and an overall similarity is calculated from the six similarities calculated by the matching unit 34, and a It is comprised of a determination unit 35 that compares with a threshold value and determines personal verification.

Reference numeral 5 denotes a control section that controls each of the above-mentioned sections and means.

FIG. 7 is a hardware configuration diagram showing an embodiment in which the personal verification device according to the present invention is configured using a microcomputer.

11 is a ten-key league constituting a personal code input section. A monochrome CCD camera 21 constitutes the face image input section 2. Reference numeral 100 denotes the interface section 31, preprocessing section 32, area extraction section 33, matching section 34, and determination section 35.
, a microcomputer that constitutes the personal characteristic storage section 4 and the control section 5, and a CPt150 that performs various processing operations.
, a program memo 'J (RO! 51) that stores the verification processing program described later, a data memory (RAM > 40) that stores personal characteristic information, an A/D converter (not shown), and a monochrome CCD. camera 21
It is composed of a frame memory 41 that A/D converts a face image captured and input into digital information and stores it, and an interface 31.

Next, referring again to the flowchart shown in FIG. 1, the operation of the personal verification device shown in FIGS. 6 and 7 will be explained.

It is assumed that the registration process in step 10 of FIG. 1 has already been completed.

First, a user who performs personal verification enters a personal code specific to the user from a personal code input unit 1 comprised of a numeric keypad reader 11 at a predetermined position. This personal code is set in advance for each individual, and personal characteristic information for each personal code is stored in advance in the personal characteristic storage section 4 of FIG. 6 or the data memory 40 of FIG. 7. When the personal code is input, the control unit 5 composed of the CPυ 50 and the program memory 51 compares it with the personal code stored in the personal characteristic storage unit 4 and determines that the personal code is normal if it is a registered personal code. The user is identified and a light (not shown) is illuminated on the user's face. At the same time, at the timing when the light reaches the maximum amount of light, the camera of the face image input unit 2 composed of the monochrome CCD camera 21 is released to capture a front face image of the user. The captured front face image information is input to the interface 31, A/D converted by an A/D converter (not shown), and stored in the frame memory 41 in FIG.
is temporarily stored. The A/D converted face image information is C
The data is input to a preprocessing section 32 that includes a PU 50 and a program memory 51. The preprocessing unit 32 performs, for example, the above-mentioned 5 obel 0 perator process in order to extract the contour image of the input front face image. Next, in order to remove random noise generated during imaging and to ensure stable matching, for example, the Median Filter process described above is performed. Further, in order to make the size of the face image constant, normalization processing such as enlargement, reduction, rotation, etc. is performed.

Next, the face image information processed by the preprocessing unit 32 is CP+1
50 and a program memory 51. The area extraction unit 33 uses shape information of the eyes and eyebrows inputted in advance, such as shape information such as horizontal elongation, left-right symmetry, continuity, etc., to the preprocessing unit 32.
Based on the processed contour face image information, both eyes and both layers are sequentially identified, and four rectangular areas containing one eye and one eyebrow are extracted (see FIG. 4). Next, the rectangular area including the eyes and eyebrows is designed to include feature points of the inner and outer corners of the eyes, and the temporal shape, that is, ΔA, BB, CC, and DD in FIG.

A feature extraction region characterized by boundaries of rectangular regions indicated by EE and FF is set. Here, the inner corners of both eyes, the outer corners of both eyes, and
The reason why the shape of the double-layered head was selected as a feature point was based on the research conducted by the present inventors, as compared to other facial features such as the shape of the nose and the area of the eyes and mouth, it was determined that the imaging conditions of each individual This is because we found that it is stable with little variation due to factors, and that it expresses individuality well.

Each of the feature extraction regions consists of, for example, β=60X j =40=2400 pixels as shown in FIG.

Next, the control section 5 composed of the CP+150 and the program memory 51 selects the six individuals inputted and stored in the personal characteristic storage section 4 in advance based on the personal code inputted in the personal code input section 1. The feature information is read out and input to the collation section 34. These six pieces of personal characteristic information are the shape information of the inner corners of the eyes, the outer corners of the eyes, and the head of both eyes, as shown in Figure 3.
It is stored as vector information of X n elements. The method of storing the settings in the personal characteristic storage section 4 will be described in detail later.

Here, each element constituting each vector is, for example, 8
This is pixel density information of 256 gradations from 0 to 255 expressed in bits.

Next, the matching 34 calculates the degree of similarity based on the first equation.

Then, the maximum one of these partial similarities p is determined as the similarity of the feature extraction region A of the inner corner of the right eye.

Similarity A=Max (similarity p) Similar operations are performed for the other five feature extraction regions to calculate the similarity of each feature point.

Next, the six pieces of similarity information calculated by the matching unit 34 are input to the determining unit 35, and the average similarity of the six pieces of similarity is calculated and taken as the overall similarity of the face information of the operator.

The determination unit compares the overall similarity with a determination value that is preset and stored for each individual, for example 0.98, and if the overall similarity is greater than or equal to the determination value, determines that the user is the person who has registered in advance. . Then, if the overall similarity is less than the determination value, it is determined that the person is a different person.

Here, a method for inputting and storing personal characteristics into the personal characteristics storage section 4 will be explained. An individual who performs registration uses the face image input unit 2 to capture a face image multiple times (for example, 10 times) at different times. Then, the preprocessing section 32 and the region extraction section 33 set six feature extraction regions for this plurality of face image data in the same manner as described above, and take the average for each feature extraction region, and then create an mxn element centered on the feature point. This vector information is set and stored for each personal code as personal characteristic information.

When this personal verification device is used as an access control device, it is connected to an electric lock, etc., and when the determination section 35 determines that the person is the person, the electric lock, etc. is unlocked under the control of the control section 5 etc. do it.

Further, in this embodiment, the personal code input unit 1 is configured with a numeric keypad, but other means such as a magnetic card reader or the like may be used as long as the code can be input.

In addition, in this embodiment, as the degree of similarity, the cosine ([
O3) was used, but the method is not limited to this and other known methods may be used.

In addition, in this embodiment, the threshold value used for determination is a constant value, for example, 0.98, but an optimal threshold value may be set for each individual to determine the individual vF! The personal code may be stored in the storage unit 4 together with the personal code.

Furthermore, in this embodiment, there are six types of personal characteristic information, but
It goes without saying that other information such as the distance between the eyes, the distance between the eyebrows, etc. may be used as additional information.

〔Effect of the invention〕

According to the present invention, personal identification can be performed with an extremely high matching rate by simply comparing the shapes of 4 to 6 feature points with pre-stored individual feature information from a frontal image of a human face, which is relatively easy to capture. can be realized.

In addition, the method for comparing and determining facial images is simple, and it is easy to correct changes in input conditions for facial images, so it is suitable for automatic personal verification.

[Brief explanation of the drawing]

FIG. 1 is a flowchart of the processing of the matching method according to the embodiment of the present invention, FIG. 2 is a diagram showing the area of reference image data for registration, FIG. 3 is an enlarged view of a part of FIG. 2, and FIG. The figure shows the area of the actual image data for matching, Figures 5(a) to (C) are diagrams explaining the process of calculating similarity for matching, and Figure 6 shows an example of the present invention. FIG. 7 is a detailed diagram of a part of FIG. 6, which shows the configuration of the verification device. (Explanation of symbols) I...Personal code input section, 2...Face image input section,
3... Collating means, 4... Personal characteristic storage section, 5... Control section. ] Figure Broom diagram (b) (C) Figure

Claims (1)

[Scope of Claims] 1. An image registration step of registering in advance image data of areas in the vicinity of at least the right inner corner of the eye, the left inner corner, the right outer corner, and the left outer corner of the individual to be matched, and the individual to be matched. an image input step of optically scanning the face of the person to generate facial image data; and a right inner corner of the eye from the image data obtained in the image input step;
A matching part extraction step of extracting image data of areas near each of the left inner corner, right outer corner, and left outer corner, and comparing the image data extracted in the matching part extraction step with the pre-registered image data. The method includes a similarity calculation step of comparing and calculating the degree of similarity between them, and a determination step of determining the individual if the similarity calculated in the similarity calculation step is greater than or equal to a predetermined value. ,
Personal matching method. 2. The personal verification according to claim 1, wherein each pixel of the image data in the image registration step and the image input step is multivalued image data, and the similarity is calculated for the multivalued image data. Method. 3. The similarity calculation is performed by calculating the inner product of the vector representation of the registered image data and the vector representation of the part-extracted image data, and the length of the vector representation of the registered image data and the part-extracted image data. 3. The personal verification method according to claim 2, wherein said image data is divided by the product of the length of a vector representation. 4. Between the image registration step and the image input step,
comprising a personal code input and identification step in which a personal code for identifying a person to be verified is input, and the process proceeds to verification processing only when the input personal code is normal; the image input step; the verification site extraction step; 4. The personal verification method according to claim 1, wherein the processing in the similarity calculation step and the determination step are performed based on a personal code. 5. The image registration step further registers in advance image data of areas in the vicinity of the inner end point of the right eyebrow and the inner end point of the left eyebrow, and the matching region extraction step further registers the image data obtained in the image input step. Image data in the vicinity of the right eyebrow inner end point and the left eyebrow inner end point are also extracted from the image data, and the similarity is also determined in the vicinity of the right eyebrow inner end point and the left eyebrow inner end point. The personal verification method according to claim 4. 6. An image registration means for pre-registering image data of areas in the vicinity of at least the right inner corner, left inner corner, right outer corner, and left outer corner of the individual to be matched; an image input means for scanning to generate facial image data; and a right inner corner of the eye from the image data obtained by the image input means;
a comparison part extraction means for extracting image data of areas near the left inner corner, right outer corner, and left eye corner; and comparing the image data extracted by the comparison part extraction means with the pre-registered image data. comprising: a similarity calculation means for comparing and calculating the degree of similarity between the two; and a determination means for determining the individual when the similarity calculated by the similarity calculation means is greater than or equal to a predetermined value; Personal verification device. 7. The personal verification device according to claim 6, wherein each pixel of the image data in the image registration means and the image input means is multivalued image data, and the similarity is calculated for the multivalued image data. . 8. The similarity calculation is performed by calculating the inner product of the vector representation of the registered image data and the vector representation of the part-extracted image data, and the length of the vector representation of the registered image data and the part-extracted image data. 8. The personal verification device according to claim 7, wherein the personal verification device performs the verification by dividing the obtained image data by the product of the length of the vector representation. 9. Enter the personal code to identify the person to be verified,
The method further includes a personal code input and identification step in which the input personal code is shifted to a matching process only when the input personal code is normal, and the image input process, the matching part extraction process, the similarity calculation process, and the determination process are performed using the personal code. The personal verification device according to any one of claims 6 to 8, characterized in that the personal verification device performs the verification based on the following. 10. The image registration means further registers in advance image data of regions in the vicinity of the right eyebrow inner end point and the left eyebrow inner end point, and the matching region extraction means further registers image data in the vicinity of the right eyebrow inner end point and the left eyebrow inner end point, and the matching part extraction means further registers the image data obtained by the image input means. Image data in the vicinity of the right eyebrow inner end point and the left eyebrow inner end point are also extracted from the image data, and the similarity is also determined in the vicinity of the right eyebrow inner end point and the left eyebrow inner end point. The personal verification device according to claim 9.