Face authentication method and face authentication device

【0001】

[Technical field to which the invention belongs]

The present invention relates to a face recognition method for inputting a face image of a person viewed from the front, calculating feature quantities such as eyes, nose, and mouth from the input image data to perform personal authentication, and a face recognition device.

【0002】

[Conventional technology]

Generally, in the face recognition method, a face image is input by a camera or the like, a face area is extracted from the input image, face parts such as eyes, nose, and mouth are extracted, and after calculating the feature amount of each part, the face recognition method is performed. Generally, the obtained feature amount is compared with the comparison target to determine whether the person is the person or another person.

【0003】

Further, as a method for extracting feature points at the time of face recognition, image data in the vicinity of the inner corner of the right eye, the inner corner of the left eye, the outer corner of the right eye, and the outer corner of the left eye is used as the feature data of the face. Refer to Patent Document 1 [0004]

Alternatively, in Patent Document 2, as facial feature points, head apex, right eyebrow left end point, right eyebrow right end point, right eyebrow upper end point, right eyebrow lower end point, left eyebrow left end point, left eyebrow right end point, left eyebrow upper point. End point, left eyebrow lower end point, right eye left end point, right eye right end point, right eye upper end point, right eye lower end point, left eye left end point, left eye right end point, left eye upper end point, left eye lower end point, nose left end point, nose right end point, nose lower end point , Lip left end point, lip right end point, lip upper end point, lip center point, lip lower end point, nose left profile contour point, nose right profile contour point, lip left profile contour point, lip right profile contour point, jaw left profile contour point , The right profile contour point of the eyebrow, and the lower end point of the eyebrow are disclosed.

【0005】

Further, in Patent Document 3, a search range for extracting the mouth, left and right eyes, eyebrows, facial contours, etc. is set for each feature portion based on the coordinates of the left and right eyes and the mouth of the face. , Techniques for extracting feature points in each search range are described.

【0006】

[Patent Document 1]

Patent Gazette No. 2690132 [Patent Document 2]

Japanese Unexamined Patent Publication No. 8-77334 [Patent Document 3]

Japanese Unexamined Patent Publication No. 9-6964 [0007]

[Problems to be Solved by the Invention]

Conventionally, when comparing face images to check whether those images are the same person image or different people, a method of detecting the feature point coordinates of the face image and comparing the feature amounts of the detected coordinates is adopted. ing.

【0008】

However, in the above technique, only the values of specific points such as eyes, nose, and mouth of the face image data are obtained, but these points are located differently depending on the individual and include instability factors. Further, in the method of setting a large number of feature points, it takes a lot of time to search for the feature points in proportion to the number of feature points, and the detection accuracy of the feature points deteriorates as the number of feature points increases.

【0009】

The present invention is characterized in how to obtain image data. That is, by using a stable part with few fluctuation factors in identifying an individual, the feature point coordinates are calculated for the same part of the face for the same person, and the feature point coordinates are different for another person. Due to individual differences in the coordinates of the feature points, differences occur in each feature amount, and the accuracy of personal authentication is improved.

【0010】

That is, the present invention is a method for solving a problem that it takes a lot of time to extract feature points and it is difficult to obtain accurate coordinates, and other feature points are calculated from the minimum feature point extraction. By doing so, the authentication accuracy is improved by obtaining many feature point coordinates in a short time.

【0011】

[Means for solving problems]

In the present invention, it is obtained by the feature point coordinate extraction unit in the image input step of inputting the face image data, the feature point coordinate extraction step of extracting the feature points from the input face image data, and the feature point coordinate extraction step. It has a feature point coordinate calculation process that calculates new feature points from several feature points, and can be obtained by calculating the coordinates of other feature points using the positional relationship and distance of the feature points. This is a method of increasing the coordinates of difficult feature points.

【0012】

Then, each image data of the feature points acquired by the feature point coordinate extraction and the feature points acquired by the feature point coordinate calculation unit is filtered to calculate the feature amount, and the obtained feature amount and the registered comparison target are calculated. It compares the features in the image data and determines whether or not they are the same person.

【0013】

For example, by detecting the positions of the right eye, left eye, and nose, and using the inner and outer division points at those three points, any point on the face is determined as a feature point, and by doing so, the same person. Then, the feature point can be placed in the same part even if it is enlarged / reduced / rotated.

【0014】

In this way, the coordinates of the new feature points can be calculated from the three feature points obtained using their distance and positional relationship, so the time required to search for the feature points can be reduced and the accuracy can be improved. it can.

【0015】

BEST MODE FOR CARRYING OUT THE INVENTION

In the present invention, a plurality of feature points (coordinates of a specific part in the face) are first obtained by a feature point coordinate extraction unit with respect to a face image (data) taken by using a CCD camera or the like. From the obtained several feature points, one polygon is formed by using the positional relationship and distance of the feature points, and geometrically as the inner and outer division points at the vertices of the generated polygon. A new feature point is obtained by expanding to (calculated by the calculation unit). This method can increase the coordinates of feature points that are difficult to obtain.

【0016】

For example, the coordinates will be described with reference to FIG. The feature points (coordinate points) are the three points where there is little movement due to facial expressions (muscles), the left end of the right eye, the right end of the left eye, and the center of the lower end of the nose. The positions of the left end A of the right eye, the right end B of the left eye, and the center C of the lower end of the nose are detected from the face image data. Based on these three points, the formed triangle ABC is developed in a plane. That is, the extracted three points of the right end A of the right eye, the right end B of the left eye, and the center C of the lower end of the nose are used as reference points, and the other feature points are obtained by vector calculation. That is, the coordinates (A, B, C) are vectorically multiplied by 1, m, and n. 1x the coordinates of point A, m times the coordinates of point B, n times the coordinates of point C, and the coordinates obtained by (1 A + m B + n C) are new. It is a feature point.

【0017】

Then, the image data at the plurality of coordinate feature points is filtered, the feature amount is calculated by the calculation unit, and the feature amount is recorded as an individual feature amount. The next time this recorded data is authenticated, the individual's feature amount is acquired by the same method, and the data is compared with the recorded recorded data by the comparison unit to see if they are the same person. Judge whether or not.

【0018】

Thus, in this embodiment, when identifying an individual, a plurality of other features are based on the minimum three points (the left end of the right eye, the right end of the left eye, and the center of the lower end of the nose) that are stable (less variable factors). Since the points are determined and the feature amount is obtained from these image data, the image data is relatively stable and the face recognition rate is high. Then, the inner and outer division points of the three points are calculated, and any point on the face is determined as a new feature point. In this way, by calculating the coordinates of the new feature points from the obtained feature points of the three reference points using their distances and positional relationships, it is possible to reduce the time required to search for the feature points and improve the accuracy. ..

【0019】

Next, an example of a method for calculating feature points will be shown. FIG. 1 is a control flowchart, and FIG. 2 is an explanatory diagram of feature points.

【0020】

First, a face image taken with a CCD camera or the like is input. (Fig. 1 (S02)) After that, the coordinates of the left end of the right eye (hereinafter referred to as the right eye) A, the right end of the left eye (hereinafter referred to as the right eye) B, and the central part of the lower end of the nose (hereinafter referred to as the nose) C are extracted from the input face image data. (Fig. 1 (S03)) [0021]

Next, for the three points that serve as the reference for the extracted right eye A, left eye B, and nose C, the calculation unit sets the coordinates of point A to 1 times, the coordinates of point B to m times, and the coordinates of point C. Multiply it by n, and use the coordinates obtained by (l, A + m, B + n, C) as the new feature points. (Fig. 1 (S04)) [0022]

In FIG. 2, the points indicated by Δ are the feature points obtained by the feature point coordinate extraction unit, and the points indicated by are the feature points obtained by the calculation unit. The parameters (1, m, n) used to calculate the feature points in Fig. 2 are (1, m, n) = (1,0,0), (0,1,0), (0,0,1). , (2, -1,0), (-1,2,0), (1, -1,1), (-1,1,1), (0, -1,2), (-1, 0,2).

【0023】

The feature amount is calculated from the data obtained by filtering the image data of each of the nine feature points obtained by the above method. (Fig. 1 (S05)) When the above feature amount calculation is completed for the two images of the comparison image and the recorded comparison target image, the feature amount comparison unit (Fig. 1 (S07)) finds the similarity of each feature point. , Compare with a predetermined threshold value and output the judgment result of whether or not the person is the same person. (Fig. 1 (S08)) [0024]

As explained above, since this face recognition is based on the minimum stable point (3 points) for identifying an individual, the recognition rate is high.

【0025】

For example, if you have to find 10 feature points and the probability of finding one point correctly is 0.95%, the probability of finding all 10 feature points correctly is about 0.60%. Assuming that there are three feature points as in the present invention, the probability that the three points can be found correctly is about 0.86%.

【0026】

In this example, the feature amount is calculated from a total of 9 points, 3 reference points and 6 feature points, but this example is an example that makes the explanation easy to understand, and other feature points can also be used. Good.

【0027】

Next, another embodiment of the feature point calculation method will be described. This embodiment shows a case where the same person can place feature points in the same part even if they are rotated.

【0028】

FIG. 3 is an explanatory diagram of an apparatus showing an embodiment of the present invention, and FIGS. 4 and 5 are explanatory views of feature points. First, a face image is input using a face image input device (Fig. 3 (S11)) such as a CCD camera. After that, the coordinates of the right eye A, the left eye B, and the nose C are extracted from the input face image data. (Fig. 3 (S12)) [0029]

Next, the point D obtained by rotating the point A counterclockwise by θ degrees around the point C is obtained by the feature point calculation device (Fig. 3 (S13), Fig. 4) and used as a new feature point. The image data of the feature points A, B, C, and D obtained by the above method is filtered, and the feature amount is calculated by the feature amount calculation device (FIG. 3 (S14)). The calculation result of the feature amount of the comparison image and the feature amount of the comparison target person registered in the feature amount database (Fig. 3 (S16)) are similar to each feature point by the feature amount comparison device (Fig. 3 (S15)). The degree is calculated, compared with a predetermined threshold value, and the determination result of whether or not the person is the same as the registered comparison target person is output. (Fig. 3 (S17)) [0030]

This example shows that the same person can place feature points in the same part even if they are rotated.

【0031】

Further, even if the face image is enlarged or reduced as shown in FIGS. 5 (a) and 5 (b), the feature points obtained by the calculation are obtained because the positional relationship and the distance relationship between the reference feature points are used. , The same person can determine the feature points in the same place. In addition, since the feature point coordinates are calculated from the coordinate relationship of the reference feature points, even if the face image is tilted as shown in Fig. 5 (b), the new feature points obtained by the calculation are the same person. Then, the coordinates of the same part can be obtained.

【0032】

[Effect of the invention]

In the present invention, the same person can newly calculate the feature point coordinates using the feature point coordinates that are relatively easy to check, calculate the feature amount for each feature point coordinate position, and compare the feature amounts. You can recognize whether you are a different person or a different person, and the probability is high. In the conventional method, a lot of time was required to extract the feature points in proportion to the number of feature points, but by using the present invention, only three feature points that are relatively easy to investigate can be extracted, and other feature points can be extracted. Is derived by calculation, so feature points can be extracted in a shorter time than the conventional method.

【0033】

Further, as the number of feature points increases, the accuracy of feature point extraction deteriorates, but if the present invention is used, the number of feature points is small, which leads to improvement in the accuracy of feature point extraction.

【0034】

Then, since the other feature points are calculated by using the distance relationship of the feature points of the individual, the feature point coordinates peculiar to the individual can be obtained. Similarly, the feature points can be placed in the same portion even if they are enlarged / reduced / rotated. For example, in one person, the coordinates of the calculated feature points are on the mouth, but in another person, they may come on the chin. In this way, since other feature points including the relationship between the position and distance of the individual's eyes and nose are extracted, it is possible to extract different feature points for each individual, which leads to improvement of the system for determining the same person by comparing feature amounts.

【0035】

As described above, by using the present invention, it is possible to construct a face recognition system with higher accuracy than the conventional method without spending time on feature point extraction.

[Simple explanation of drawings]

FIG. 1 is a flowchart showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram of feature point positions on a facial image showing an embodiment of the present invention and a method of arranging the feature points.

FIG. 3 is a device configuration diagram showing an embodiment of the present invention.

FIG. 4 is an explanatory diagram of feature point positions on a face image showing an embodiment of the present invention and a method of arranging the feature points.

FIG. 5 is an explanatory diagram of feature point arrangement in reduction and rotation of a face image showing an embodiment of the present invention.

[Explanation of symbols]

A Feature point (left end of right eye) B Feature point (right end of left eye) C Feature point (center of lower end of nose) D Calculated feature point