JPH06295331A - Personal recognition device - Google Patents

Abstract

(57) [Abstract] [Purpose] The present invention aims to identify an individual by a face image. [Structure] A position indicating unit 4 for inserting a specific portion of a face into a predetermined frame of the image input unit 1, a position detecting unit 5 for detecting the position of the face portion, and a face from the position of the specific portion of a face. Feature extraction means 10 for extracting features up to other parts of
Individuals are identified from the feature registration unit 11 that registers the features in advance and the face determination unit 12 that determines and compares the extracted features with the features that are stored in advance.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a personal recognition device for detecting a human face in an image and recognizing a specific individual from the detected face.

[0002]

2. Description of the Related Art A conventional personal identification device of this type has been used in entry / exit management and security fields to perform personal identification by image recognition of fingerprints and identification of blood vessels of retinas and image recognition of signatures. Came.

[0003]

However, the method of the above-mentioned conventional example is not sufficiently prevailing, for example, because the detection of fingerprints by image recognition has social resistance and anxiety about privacy. Generally, in the conventional fingerprint detection, an image of a fingerprint is taken, and this image is preprocessed and binarized. After that, the ridge end points, bifurcation points, and ridge direction extraction are performed,
It is a general method to compare the input pattern with the registered pattern from a part or all of this information. However, with this method, the end points and branch points of about 100 ridges are checked, which is relatively expensive and has been used only in some companies and in places with extremely high secrecy.

Further, in order to recognize the blood vessel pattern of the retina, the image of the inside of the eye is picked up by applying the eye to a place like a peep window.
It is popular because it is annoying to look into the viewing window, there is anxiety that bad light rays may not hit the eyes (although it does not actually hurt), and it is more expensive than fingerprint detection. Not not.

On the other hand, there is recognition of a signature that detects a handwriting instead of a part of the body, but at the present time when character recognition is also started, it is said that it is necessary to sign every time rather than a technical problem. The annoyance hindered the spread. Signing is generally performed at the time of giving and receiving money, but there is a problem that it is inconvenient to use when signing every time when opening and closing the door.

Face recognition has also been studied, but in the vertical direction the upper part of the hair of the head to the chin, and in the horizontal direction both ends of the hair,
Many of them aim to improve the recognition rate by obtaining eye and nose positions from the contours of both ends of the face and analyzing by pattern matching and brightness comparison using a large computer. There was no practical one that could identify non-family members. What is not simple and practical is (1) The calculation capacity becomes large in order to perform matching of the entire face, and a large capacity memory and high speed CPU must be used. Further, (2) image processing for recognizing the position of a face, for example, eyes, nose, mouth, etc. is complicated. That is, since the positioning for recognizing each part of the face is a method of obtaining the position of each part from the outer contour of the face, it is uncertain when the hairstyle changes (for example, haircut) or when the mouth is open. The algorithm becomes complicated because various elements are included. There was a problem such as.

The present invention solves the above-mentioned problems, and stands in front of a personal recognition device so that a specific part of a face is placed at a predetermined position, and eliminates others other than those registered by detecting features of several parts of the face. The first purpose is to be able to do.

A second object is to enable simple feature detection by aligning the approximate position of the face by looking at one's face using a mirror.

A third object is to easily detect a feature by aligning the face by bringing the eyes to a position where the two light emitting elements can be seen.

A fourth object is to prevent erroneous recognition by performing fuzzy inference on features.

[0011]

In order to achieve the first object, the present invention mainly puts an image inputting means for picking up a face and a specific portion of the face in a predetermined frame of the image inputting means. A position indicating section for detecting the position of the face portion from the face image contained in the position indicating section,
A feature extracting means for extracting features from the position of the face detected by the position detecting means to other parts of the face; a feature registering means for pre-registering the features obtained from the feature extracting means;
Face determining means for determining the characteristics extracted by the characteristic extracting means and the characteristics stored by the characteristic registering means are provided.

Further, in order to achieve the second object, in addition to the above-mentioned constitution, the present invention is provided with a mirror, and the position indicating portion is a mirror for aligning a specific portion of the face projected on the mirror within a predetermined frame. It is provided above.

Further, in order to achieve the third object, in addition to the above-mentioned structure, the present invention is such that two light emitting elements are made incident corresponding to the left and right eyes.

In order to achieve the fourth object, according to the present invention, the face determination means includes a control rule storage unit for storing a function of certainty from data of the feature registration unit, an output of the feature extraction unit, and a control rule storage unit. A fuzzy inference unit for performing fuzzy inference based on the output is provided.

[0015]

With the above-described structure, the personal recognition apparatus of the present invention captures a face of a person by the image input means while looking at the position pointing unit, and allows a specific portion of the face, for example, the eye portion to enter the position pointing unit. Image input means is installed, the position of the eyes is detected from the input image by the position detection means, and the detected position and other parts of the face, for example, the position to the nose and mouth are used as distance information or as a brightness distribution. Extracted by the feature extraction means,
Individual recognition is performed by comparing and determining the features previously extracted by the feature registration means by the same method and the features during use by the face determination means.

Further, the position indicating section is provided on the mirror, and the eye position is easily specified by placing the eye position at a predetermined position in the image by looking at the position indicating section of the mirror. Perform similar personal recognition.

Furthermore, the position indicating portion can be easily specified by the structure in which the left and right eyes of the two light emitting elements see each other, and the position of the eyes can be easily specified, and the individual recognition is performed as in the first method.

Then, when face determination is performed, feature data is fuzzy inferred using a certainty function stored in the control rule storage unit to perform face determination.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

In FIG. 1, 1 is an image inputting means, 2 is a front of the image inputting means 1, and a light incident path is a half mirror section 3 so that light enters the image inputting means 1. Reference numeral 4 denotes a position indicating portion provided on the mirror 2, and is provided at a position such that when a person looks at the position indicating portion 4 from the front, the position of the eyes enters the position indicating portion 4. Reference numeral 5 designates the image memory unit 6 and a luminance distribution detecting unit 7 for extracting the luminance distribution of a specific region of the image memory unit 6, a portion determining unit 8 for determining a face portion from the extracted luminance distribution, and a position of the determined portion. The position detecting means constituted by the position specifying section 9 is a feature extracting means for extracting the distance and direction of the position of each part of the face, 1
Reference numeral 1 is a feature registration means in which the values extracted by the feature extraction means 10 are registered in advance, and reference numeral 12 is a face determination for comparing the feature registered in the feature registration means 11 with the feature at the time of identification. It is a means.

FIG. 2 is a view in which the personal identification device 13 is installed near the door 14. Reference numeral 15 is a confirmation lamp when personal recognition is confirmed, 16 is a start switch of the personal recognition device, 17
Is a person and 18 is the person's face. Further, a face 18 is projected on the mirror 2, and the position indicating unit 4 is placed on the mirror 2.
There are lines 19 and 20 which are

Next, the operation of the configuration of this embodiment will be described. FIG. 3 is a flowchart showing this operation, and FIG.
FIG. 4 is a model diagram for explaining face recognition. A case where the personal identification device 13 is used as an entrance management device for the door 14 will be described. People with a security level that can enter the room first register their characteristics. First, registration will be described.

The person 17 mirrors the face 18 for registration.
So that the position of the eyes of the face is located between the lines 19 and 20 which are the position designating unit 4 (step 10).
1). Then, the start switch 16 of the personal identification device 13 is pressed (step 102), and the mirror 2 is pressed in step 103.
An image of the face 18 is input by the image inputting means 1 on the back side of and is stored in the image memory unit 6. FIG. 4A is a binarized face image obtained by taking edge portions from the stored face image. Here, 21 is eyes, length a is the distance between eyes, length b is the width of the face at the height of eyes 21, length c is the distance between eyes 21 and nose 22, and length d is the distance between eyes 21 and mouth. 23 is the distance from the upper part. In step 104, horizontal scanning is performed to search the position of the eye 21 from the image stored in the image memory unit 6 in the image entered in the position designating unit 4, and the binary brightness level 1 in the position designating unit 4 is changed. The distribution is detected by the brightness distribution detector 7. FIG. 4B is a brightness distribution in the position designating section 4. In step 105, the partial determination unit 8 determines the eyes 21 that are one of the face portions from the obtained luminance distribution. Since the positions of both ends of the face have sharp edges, the number detected is large (sharp peaks at both ends on the graph). It can be seen that there is a broad distribution inside the edge of the face, and this is the portion of the eye 21. The position specifying unit 9 calculates the center of each distribution of the eyes 21 to obtain the position of the eyes 21. Figure 4
In (B), length a is the distance between the eyes 21, and length b is the width of the face. Next, in step 106, the position specifying unit 9 obtains the position of the nose 22 with the distance in the vertical direction from the position of the eye 21 as c, and in step 107, the distance d above the mouth 23 is also calculated by the position specifying unit 9. Ask. In step 108, using the distances a, b, c, d previously obtained by the position detecting means 5, c / a, d / a are obtained by the feature extracting means 10 and calculated as the facial features of the person, and the feature registering means 11 Step 10
Store at 9. As for the registration method, c / a is classified into 7 levels and d / a is classified into 7 levels, and the registration is performed at the applicable position in the 7 * 7 matrix arrangement. FIG. 5 shows the addresses of the matrix to be registered, where 3-3 is a standard person, and the address of the person just obtained is 2-4, for example. With the above, the registration work is completed, and the confirmation lamp 15 for registration completion is turned on (step 110).

Next, a method for personal identification will be described.
FIG. 6 is a flowchart showing this operation. The person 17 who is about to enter the room brings the face 18 closer to the mirror 2 so as to be recognized, and positions the eyes 21 of the face so as to be between the lines 19 and 20 which are the position indicating section 4 (step 201). After that, the activation switch 1 of the personal identification device 13
6 is pressed (step 202), and in step 203, the image of the face 18 is input by the image input means 1 on the back side of the mirror 2 and stored in the image memory unit 6. In step 204, the image stored in the image memory unit 6 is horizontally scanned in order to search the position of the eyes from the image entered in the position designating unit 4, and the distribution of the binarized brightness level 1 in the position designating unit 4 is performed. Is detected by the brightness distribution detection unit 7. In step 205, the partial determination unit 8 determines the determination of the eyes, which is one of the face portions, from the obtained luminance distribution.
Done in. The center of each of the eye distributions is set to the position specifying unit 9
Then, the position of the eye 21 is obtained. Here, as in the case of registration, the distance between the eyes 21 is a and the width of the face is b.

Next, in step 206, the position specifying unit 9 finds the position of the nose 22 from the position of the eyes 21 with the vertical length c, and in step 207, the length d to the upper part of the mouth 23 is determined. Similarly, it is obtained by the position specifying unit 9. Step 208
Then, the lengths a, b, which have been previously obtained by the position detecting means 5,
Using c and d, the feature extraction means 10 finds c / a and d / a and calculates them as the features of the person's face. Next, step 2
In step 09, the face determination means 12 sets c / a in seven stages at the time of registration, d
/ A is classified into 7 levels, and it is determined whether or not it matches a given position in the 7 * 7 matrix arrangement. If the person who is about to enter the room is detected at the address 2-4, that is, if it matches the registered data, the process goes to step 210, the door 16 is opened, and the confirmation lamp 15 is also turned on (step 211). If, for example, the detected data is the address 3-2, that is, if it does not match the registered data, the security of the room is ensured as a person who cannot enter the room without opening the door 16. And step 21
1 and the confirmation lamp 15 for recognition completion is turned on.

As a method of detecting the position of each part of the face, it is also possible to directly obtain the luminance distribution from the multivalued image and determine the position from the pattern.

As described above, the face recognition can detect a substantially constant specific position of the face from the input image when the environmental conditions at the time of inputting the image, for example, the lighting method is good, and as described above. It is only necessary to determine whether or not it is within one area of the 7 * 7 matrix.

Next, a method of reducing recognition errors by fuzzy reasoning when the environment at the time of entry and exit may change will be described.

The face determination means 12 is a control rule storage section 24 for storing a function of certainty from the data from the feature registration section 11 and an output from the feature extraction means 10 and a control rule storage section 2.
And a fuzzy inference unit 25 which performs a fuzzy inference based on the output from 4.

If the control rules registered in the control rule storage unit 24 are registered at the same registration address in the above-mentioned example,
The registered address 2-4 (Miemaru) is assumed to have the highest recognition coincidence rate, and the recognition rate is 1 (probability is 1). And, 1-3, 2-3, 3-indicated by double circles around it
The recognition rates of 3, 1-4, 3-4, 1-5, 2-5, and 3-5 are 0.8 (probability is 0.8). Furthermore, 1-2, 2-2, 3-2, 4-2, 4 indicated by single circles around it
-3, 4-4, 4-5, 4-6, 3-6, 3-7, 2-
If it is 6, 1-6, it is 0.5.

Further, using the information on the distance between the ends of the face on both sides of the eye 21 called b obtained in FIG. 4 (A), the possible values of b / a are divided into 1 to 7. Here, b / a is called a face contour. If the registered value of the face contour b / a (here, it is 4), the recognition rate of the face contour is 0.9, and if it is the values (3 and 5) on both sides thereof, the recognition rate is 0.7, and further away. If (2 and 6), the value is 0.3.

Next, the fuzzy inference unit 25 performs fuzzy inference based on the certainty of the two-dimensional axes c / a and d / a and the certainty of the one-dimensional axis b / a. Now, the fuzzy logical product of the probability obtained from the determination condition in the case of two dimensions and the probability obtained from the determination condition in the case of one dimension is obtained, and the value is 0.5 or more, that is, the probability of being half or more. In some cases, the person is recognized as a registered person in the face determination, and if it is 0.5 or less, that is, if the probability is less than half, the person is not a registered person.

For example, if the position detected on the two-dimensional axis is 1-
If it is 3, the recognition rate is 0.8, and the address of the face contour at that time is 5
If so, the recognition rate becomes 0.7, and the fuzzy AND is 0.
The person detected at 56 is determined to be a registered person. Therefore, up to now, it was possible to prevent erroneous recognition even if there was a change in the environment by including the condition of certainty in the surroundings, except that one area of the 7 * 7 matrix was judged as an error. .

The above-mentioned operation was explained by using the image inputting means to capture the image of the face and using the mirror to confirm the image to be captured, but it is also possible to confirm the input image with a liquid crystal display device. Of course it is possible.

Further, although the example in which the position indicating portion is provided on the mirror has been described, another method, for example, two light emitting elements having a narrow irradiation angle are installed separately, and a face is held at a position where both of them can be seen. By moving, the eye position can be put in the position indicating section.

In FIG. 7, two light emitting elements 26 and 27 are mounted on a mounting base 28 and partitioned by a partition plate 29.
The light emitting element 26 expands like optical paths 30 and 31, and the light emitting element 27 expands like optical paths 32 and 33. By placing the position of the eye in the shaded area 34, the left eye is illuminated by only the left light emitting element 26 and the right eye is illuminated by only the right light emitting element 27, so that the light can be seen. By doing so, the position of the face can be easily aligned, and a person who does not know this method is not registered and can be excluded before the identification by the image. The reason why the number of light emitting elements is two is that when the eyes are viewed by both eyes, each eye sees the corresponding light emitting element, so that the positional deviation between the left and right is reduced.

[0037]

As described above, in the personal recognition device of the present invention, the image input means is set so that the specific portion is included in the position indicating portion, and the position of the specific portion of the face is detected from the obtained image, The following effects can be obtained by comparing the features from the detected portion to other face portions.

(1) Since the entire face is not matched, the calculation capacity is small and the memory is small.

(2) Since the rough position of the specific portion of the face that has entered the position pointing portion is known in advance, the determination of the specific portion of the face becomes easy. Therefore, the feature between the specific parts can be easily extracted.

(3) It is possible to create a system having a degree of freedom such that many people can be identified by increasing the number of feature combinations, and family members can be identified by decreasing the number of combinations.
In particular, an inexpensive system can be created for home use.

When the number of features to be detected is increased, it becomes 1
If the two data do not match, even if the person is OK, it is determined that they are different. However, by performing fuzzy inference based on the certainty of each data, misrecognition is reduced. Further, erroneous recognition is reduced even with environmental changes (such as changes in lighting) when an image is input.

Further, since a mirror is provided in front of the image input means for inputting a face image to the image input means, an appropriate image can be taken in only by looking at the mirror so that one's face can enter. it can.

By writing the position indicating portion on the mirror, the eye position can be easily detected only by looking at the mirror so that one's eyes are aligned with the position indicating portion.

The position indicating portion can be easily specified by the structure in which the left and right eyes of each of the two light emitting elements see each other, and the person who does not know how to use can be repelled in advance.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a moving object recognition apparatus according to an embodiment of the present invention.

FIG. 2 is a schematic view of the device arranged near the door.

FIG. 3 is a flowchart for registering characteristics of the device.

FIG. 4 is an explanatory diagram showing a face image and face specific portion data of the apparatus.

FIG. 5 is a matrix layout diagram of data of the device.

FIG. 6 is a flowchart for recognizing the device.

FIG. 7 is a configuration diagram of an embodiment of a position pointing unit of the same device.

[Explanation of symbols]

 1 Image Input Means 5 Position Detection Means 10 Feature Detection Means 11 Feature Storage Means 12 Face Determination Means

Claims (5)

[Claims] 1. An image inputting device mainly for picking up an image of a face, a position instructing part for putting a specific part of the face within a predetermined frame of the image inputting device, and a face image from the face image stored in the position instructing part. Position detection means for detecting the position of each portion, feature extraction means for extracting features from the position of the face portion detected by the position detection means to the position of the other portion of the face, and the feature extraction means A personal recognition device comprising: a feature registration means for registering the feature in advance; and a face determination means for comparing and determining the feature extracted by the feature extraction means and the feature stored in the feature registration means. 2. The personal recognition device according to claim 1, wherein the image input means is behind the mirror, and the face is mainly photographed through the mirror. 3. The personal identification device according to claim 2, wherein the position pointing unit is provided on the mirror so that a specific portion of the face projected on the mirror is fitted within a predetermined frame. 4. The personal identification device according to claim 1, wherein the position indicating section is provided with two light emitting elements so as to be incident so as to correspond to the left and right eyes. 5. The face determination means is fuzzy from a control rule storage section for storing the likelihood of the feature obtained by the feature registration means, the data extracted by the feature extraction means and a function of the likelihood of the control rule storage section. The personal recognition device according to any one of claims 1 to 4, comprising a fuzzy inference unit for inferring.