JP2012128792A - Personal authentication device and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a personal authentication device and a computer program which can properly perform personal authentication even when the authentication is performed with data including an area in which feature point data is not extracted.SOLUTION: A personal authentication device comprises: division means 3 for dividing a face image into a plurality of areas; second extraction means 4 for extracting pieces of feature point data from all areas in which feature point data can be extracted; registration means 6 for registering pieces of registration feature point data of a registration face image; and collation means 7 for, for each corresponding area, collating collation feature point data of a collation face image with the registration feature point data, and based on collation results, performing collation of the entire collation face image and the entire registration face image. Based on only collation results of areas in which both the collation feature data and the corresponding registration feature point data are extracted, the collation means 7 performs the collation of the collation face image and the registration face image.

Description

  The present invention relates to a person authentication device and a computer program using an image. In particular, the present invention relates to a person authentication apparatus and a computer program for determining whether a person is a pre-registered person by comparing a face image acquired from the person with a face image registered in advance by using the person as an authentication target. .

  When performing person authentication using a person authentication device, first, in order to acquire a face image from a person (for example, a pedestrian), an imaging camera is installed toward the flow line of the person.

  Next, the face image of a person who has entered the shooting area of the imaging camera is shot at several frames per unit time, and the face image of the person in the image is detected.

  By comparing the detected face image with a pre-registered face image, it is determined whether the face image is a face image of the same person as the pre-registered person. With this operation, the person is authenticated.

  Here, when the face width of the detected face image is greater than or equal to a predetermined size, a plurality of face images can be acquired for the same person and supplied to the subsequent face matching process.

  In addition, in order to improve the authentication accuracy and the authentication speed of an acquired face image, a face image is divided into arbitrary sizes and collated for each division unit (for example, patent document) 1).

  Furthermore, for the same purpose, there is also disclosed a technique in which a face image is divided into nine areas, feature extraction is performed for each area, and collation is performed (for example, see Patent Document 2).

  Further, for example, Patent Document 3 discloses a method for determining facial feature points.

JP 2006-85249 A JP 2009-2111425 A JP-A-61-175509

  However, in the case shown in Patent Document 1, the face image is divided into arbitrary sizes, and the face image is arbitrarily divided even though there are parts such as eyebrows, eyes, nose, mouth, and cheeks. For this reason, there are cases where parts such as the eyes and the nose are divided, and there is a problem that the case where the authentication accuracy is affected can be created.

  Further, in the case shown in Patent Document 2, the face image is divided into nine areas, and feature points are extracted for each area to be collated. If you cannot see your eyebrows, cover your eyes, or use it as a face image for verification, the area will be treated as inconsistent, and even if other areas match, there will be a case where you are not authenticated. It was.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a person authentication device and a computer program capable of appropriately performing person authentication even when there is an area where feature points are not extracted.

  The person authentication device according to the present invention includes a dividing unit that performs a process of dividing the face image into a plurality of areas, and a process of extracting feature point data for each of the plurality of areas. Extraction means for performing all extraction possible areas; registration means for previously registering the extracted feature point data as registered feature point data of a registered face image with respect to a registered face image to be used as a reference for collation; The matching feature point data of the matching target face image, which is the extracted feature point data of the matching target face image to be a matching target, and the registered feature point data are matched for each corresponding area, A collation unit configured to collate the face image to be collated with the registered face image based on the collation result, wherein the collation unit includes the matching target feature point data and the corresponding registered feature point data; Based collation result only in both is extracted areas and performs collation with the registered face image and the collation target face image.

  According to the first person authentication device of the present invention, it is possible to extract the dividing unit that performs the process of dividing the face image into a plurality of areas and the process of extracting the feature point data for each of the plurality of areas. Extraction means for all areas, registration means for previously registering the extracted feature point data as registered feature point data of a registered face image for a registered face image to be used as a reference for comparison, and a face to be verified Collation means for collating image target feature point data and registered feature point data for each corresponding area, and collating the collation target face image with the registered face image based on the collation result And the collating means includes the collation target face image and the registered face based only on a collation result in an area where both the collation target feature point data and the corresponding registered feature point data are extracted. By performing the matching between the image, the registration face image with the exception of the areas that can not extract the feature point data, verification can be performed, reducing unnecessary verification can reduce the false certificate of the collation.

1 is a configuration block diagram of a person authentication device according to a first embodiment; 3 is a flowchart showing a registration operation of the person authentication device according to the first exemplary embodiment; 3 is a flowchart showing a collation operation of the person authentication device according to the first exemplary embodiment. It is a figure which shows the mode of a division | segmentation of a face image. It is a figure which shows the division | segmentation of a face image, and the data content list of feature point data. It is a figure which shows the specific example of feature point extraction. It is a figure which shows the division | segmentation of a face image, and the data content list of feature point data. It is a figure which shows the division | segmentation of a face image, and the data content list of feature point data. It is a figure which shows the division | segmentation of a face image, and the data content list of feature point data. It is a figure which shows the division | segmentation of a face image, and the data content list of feature point data. It is a figure which shows the mode of a division | segmentation of a face image.

<A. Embodiment 1>
<A-1. Configuration>
FIG. 1 shows a configuration diagram of a person authentication device according to the present invention. Note that the person authentication device can be realized by software, and the software can be installed in a computer and executed, thereby causing the computer to function as a person authentication device.

  As shown in the figure, the person authentication device includes an image pickup unit 1 that picks up an image in an image pickup region, and a first extraction unit that extracts a person face image (a face image to be collated) from the person image acquired by the image pickup unit 1. 2 and a dividing unit 3 that performs processing for dividing the data of the face image to be collated into a plurality of areas. In the present invention, it is only necessary to acquire face image data that can be divided by the dividing means 3, and means for acquiring a person image and means for extracting a face image from a person image are not indispensable configurations. That is, the image pickup unit 1 and the first extraction unit 2 do not necessarily have to be provided when data of face images (a face image to be collated and a registered face image to be described later) is provided from a separate unit. Here, the divided areas in the dividing means 3 can be arbitrarily set, for example, corresponding to each part (eye, nose, mouth, etc.) that is a feature of the face so that feature point data described later can be easily extracted. It is desirable to set it.

  Further, a second extraction is performed for extracting feature points for each divided area of the face image to be collated as feature point data (particularly, feature point data extracted for the face image to be collated is used as collation target feature point data). Means 4 (extraction means) is provided. The extraction process is performed on all areas where extraction is possible. Here, the feature points refer to positions, sizes, thicknesses, angles, and the like of the outer shape of the face, eyebrows, eyes, nose, ears, mouth, and the like, and their relative relationships (see, for example, Patent Document 3). Extraction of feature point data for each area is to measure and extract the position, size, thickness, angle, etc. of facial parts (face outline, eyebrows, eyes, nose, ears, mouth, etc.). That is.

  The feature point data extracted by the second extraction unit 4 is checked by the check unit 5 to determine whether there is an area in which the feature point data could not be extracted depending on the state of the image (or the imaged person).

  In the case of the collation operation, the collation unit 7 obtains the collation target feature point data of the obtained collation target face image and the registered feature point data of the registered face image registered in advance in the registration unit 6 for each corresponding area. Matched. At this time, the area for which it is determined by the check means 5 that feature point data could not be extracted is excluded from collation targets.

  Here, the registered face image in the registration means 6 is a face image to be used as a reference for collation, and is similar to the face image to be collated and is obtained by extracting a face image from the acquired person image. The registered feature point data in 6 is obtained by extracting feature point data for each divided area. The divided areas are set so as to correspond to the matching target feature point data and the registered feature point data.

  Based on the result of the verification, the verification target face image and the registered face image are verified, and if it is determined that the face person displayed in the verification target face image matches the person registered in advance, a notification is sent. Notification can be made by means 8.

<A-2. Registration operation>
FIG. 2 is a flowchart showing a registration operation of a registered face image and registered feature point data.

  First, a person image is acquired using the imaging unit 1, and a face image is extracted by the first extracting unit 2 and input to the dividing unit 3 (step S11).

  In the dividing unit 3, the processing target area of the input face image is subjected to termination structure / normalization (step S12). Further, the face image is divided into five areas (step S13), and the second extraction means 4 extracts feature point data for each area (step S14). As specific contents of extraction, the position, size, thickness, angle, and the like of facial parts (for example, the outer shape of the face, eyebrows, eyes, nose, ears, mouth, etc.) are measured in each area.

  FIG. 4 shows five area divisions of the face image processing target area. FIG. 4A shows an image obtained by terminating and normalizing the processing target area after inputting the face image. FIG. 4B shows the result of dividing this into the left eyebrow, right eyebrow, left eye, right eye, and nose / mouth / cheek area under the eye. Further, for each area, A area (left eyebrows), B area (right eyebrows), C area (left eye), D area (right eye), and E area (nose / mouth / cheek) are set ( FIG. 4 (c)).

  Acquired face images include those in which the hair is applied to the eyebrows and those in which the hair is applied to the eyes. In such a case, for example, when measuring the position, size, thickness, angle, etc. of the eyebrows, it is impossible to measure the position, size, thickness, angle, etc. of the eyebrows because the shape of the eyebrows cannot be grasped. It becomes. For this reason, the feature point data may not be extracted from the face image divided into each area. Thus, the check means 5 checks whether or not there is an area where feature point data could not be extracted (step S15 in FIG. 2).

  FIG. 5 shows a data content list of feature point data. FIG. 5A shows the area of the face image corresponding to FIGS. 4B and 4C, and FIG. 5B shows the extracted feature point data corresponding to the face image. It shows a pattern.

  A specific example of a case where there is feature point data that is not extracted as shown in FIG. 5B is shown in FIG. For example, when the left eyebrow is covered with hair, as shown in FIG. 6 (a), the boundary above the eyebrows becomes invisible and it is determined that feature point data cannot be extracted.

  If the left eye is covered with hair, as shown in FIG. 6 (b), the eyeball portion of the eye is obscured, and the eyelid, the pupil / eye contour cannot be obtained, and the feature point data is obtained. It is determined that it cannot be extracted.

  The second extraction means 4 excludes such areas and extracts feature point data of the entire face image composed of areas from which feature point data can be extracted (step S16). The feature point data for the entire face image is extracted only by the feature point data for each area, but each part of the face can be measured, but between the eyes and between the eyes and the nose, This is because the relative relationship between the parts cannot be measured. By extracting feature point data from the entire face image, the relative relationship within and between parts is measured along with the measurement of each part of the face.

  FIG. 6C is an image diagram of a case where, for example, the left eyebrow cannot capture feature point data. FIG. 6D is an image diagram of a case in which the right eyebrow fails to obtain feature point data, for example.

  Here, the registration operation for the area from which the feature point data is extracted is shown.

  When the registered feature point data of the registered face image has been extracted for all areas, the matching target feature point data of the corresponding face image to be matched may be extended to the eye or applied to the entire area. May not be extracted.

  In such a case, as the feature point data used for collation, only the area extracted from both face images, that is, the area from which no feature point data is extracted from one face image, is identified from the collation target. exclude.

  Therefore, as the feature point data of the registered face image, an exclusion pattern of the feature point data is created in advance in order to correspond to the face image to be collated with various extraction patterns (step S17). That is, first, feature point data is registered in all extracted areas (if all areas are extracted, they are registered for all areas), and sequentially registered as feature point data for each area could not be extracted. To go. Registration is performed in the registration means 6 (step S18).

  A pedestrian whose left eyebrow area is hidden by hair at the time of registration does not always walk with the left eyebrow area hidden by hair even at the time of verification. If the wind blows or the hair is shortened, the area where feature point data cannot be extracted changes.

  In order to cope with such a case, an exclusion pattern in which feature point data of various areas could not be extracted is created.

  By registering in this way, as shown in FIG. 5B, the extraction pattern of registered feature point data can be used properly according to the extraction pattern of matching target feature point data.

<A-3. Verification operation>
FIG. 3 is a flowchart showing the collation operation of the collation target face image and the collation target feature point data.

  First, a person image is acquired using the imaging unit 1, and a face image is extracted by the first extracting unit 2 and input to the dividing unit 3 (step S21).

  In the dividing unit 3, the processing target area of the input collation target face image is subjected to termination structure / normalization (step S22). Further, the face image is divided into five areas (step S23), and the second extraction means 4 extracts feature point data for each area (step S24). As in the face registration, the position, size, thickness, angle, and the like of face parts (for example, face outline, eyebrows, eyes, nose, ears, mouth, etc.) are measured in each area.

  The check means 5 checks whether there is an area where feature point data could not be extracted (step S25). The second extraction unit 4 excludes such areas and extracts feature point data of the entire face image composed of areas from which feature point data can be extracted (step S26).

  Here, the collation operation for the area from which the feature point data is extracted is shown.

  In some cases, an area from which feature point data is not extracted at the time of registration can be extracted at the time of matching. For example, long hair (the eyebrows are on the eyes) was registered at the time of registration, but the hair was cut and rounded (the eyebrows were not on the eyes) at the time of verification.

  For this reason, the feature point data exclusion pattern is created in order to correspond to the registered face images of various extraction patterns for the matching target feature point data of the matching target face image (step S27). That is, first, among the feature point data of all the areas that have been extracted, the feature point data of the area from which the feature point data has been extracted in the registered face image is extracted as the matching target feature point data. The feature point data of the area that has not been extracted is excluded from collation because it could not be extracted.

  When the feature point data to be collated is determined, the collation target feature point data and the registered feature point data are collated by the collating means 7 (step S28).

  As a result of the collation, if it is determined that the collation target feature point data and the registered feature point data match, that is, it can be determined that the collation target face image and the registered face image match, notification of matching is given (step S30). In the case of a mismatch, non-notification (step S31) can be made, but a mismatch notification may be made.

  Examples of matching between registered feature point data and matching target feature point data are shown in FIGS. FIG. 7A shows a case where the registered face image is normalized, and there is no area where the registered feature point data cannot be extracted. FIG. 7B shows an extracted pattern of feature point data corresponding to a face image. There are various other extraction patterns from when all the feature point data can be extracted in each area (NO.1) to when there is no left eyebrow (NO.2).

  FIG. 8A is a normalized face image to be collated, and corresponds to a pattern without left eyebrows (NO. 2) (FIG. 8B). However, as described above, the area from which the feature point data to be collated cannot be extracted is not limited to the pattern from which the registered feature point data can be extracted. Therefore, the feature point data is also extracted for other patterns. (FIG. 8B).

  Since the face image to be collated is a pattern without the left eyebrow (NO. 2), the NO. 2 and NO. 2 (feature point data list of registered face image). The collation is performed in the collation means 7 (step S28 in FIG. 3).

  Thus, by eliminating the feature point data of the area where the feature point data could not be extracted, it is possible to reduce useless collation operations. Furthermore, it is possible to reduce erroneous authentication that occurs when collation is performed using an area where feature point data is not extracted and cannot be used for collation. As a result, it is possible to contribute to the improvement of the person authentication speed.

  Further, examples of matching between registered feature point data and matching target feature point data are shown in FIGS. FIG. 9A shows a normalized registered face image in which the left eyebrows are hidden by the hair. FIG. 9B shows an extracted pattern of feature point data corresponding to a face image. Since the left eyebrow is hidden by the hair, the case where all the feature point data can be extracted in each area (NO. 1) is not created.

  FIG. 10A shows a case where the face image to be collated is normalized and there is no area where feature point data could not be extracted. The feature point data is also extracted for various patterns (FIG. 10B).

  Since the registered face image is the pattern without the left eyebrow (NO. 2), the NO. 2 and NO. 2 (the feature point data list of the face image to be collated) is collated. The collation is performed in the collation means 7 (step S28 in FIG. 3).

  Here, FIG. 11 shows division patterns of other face images (collation target face image, registered face image). FIGS. 11A and 11B are patterns in which the nose / mouth / cheek portion is divided into three. A region (left eyebrows), B region (right eyebrows), C region (left eye), D region (right eye), E region (left cheek), F region (nose / mouth), G region (right cheek) Setting is made (FIG. 11B).

  FIGS. 11C and 11D show patterns in which the nose / mouth / cheek portion is divided into four parts. A region (left eyebrows), B region (right eyebrows), C region (left eye), D region (right eye), E region (left cheek), F region (nose), G region (mouth), H region ( (Right cheek) is set (FIG. 11 (d)).

  As the number of divisions increases, the area where feature point data cannot be extracted increases, and the load for creating feature point data lists during registration and verification increases, while the area excluded from verification can be localized , Leading to improved authentication accuracy.

<A-4. Effect>
According to the first embodiment of the present invention, in the person authentication device, the dividing unit 3 that performs the process of dividing the face image into a plurality of areas, and the feature point data are extracted for all the areas that can be extracted. , Second extraction means 4 as extraction means, registration means 6 for registering registered feature point data of the registered face image, matching target feature point data of the matching target face image, and registered feature point data. A collating unit 7 that collates each area and collates the collation target face image with the registered face image based on the collation result. The collating unit 7 includes collation target feature point data and corresponding registered feature points. By collating the face image to be matched with the registered face image based only on the matching result in the area where both the data was extracted, registration and matching of the face image is performed except for areas where feature point data cannot be extracted. Can do , Reduce the wasteful verification, it is possible to reduce the false positives card in the match. In addition, the authentication speed can be improved.

  Furthermore, since the area that cannot be collated is removed from the collation target, it is possible to immediately and flexibly cope with the influence of the hair and the change in facial expression of the face authentication target pedestrian or the like.

  Further, as the number of divisions increases, the area to be excluded can be localized, leading to an improvement in authentication accuracy.

  In addition, according to the first embodiment of the present invention, the person authentication apparatus further includes the notification unit 8 that notifies only when the collation target face image matches the registered face image as a result of the collation by the collation unit 7. Thus, notification can be made efficiently.

  In the embodiment of the present invention, the material, material, conditions for implementation, etc. of each component are also described, but these are examples and are not limited to those described.

  DESCRIPTION OF SYMBOLS 1 Image pickup means, 1st extraction means, 3 Division means, 2nd extraction means, 5 Check means, 6 Registration means, 7 Verification means, 8 Notification means

Claims (7)

Dividing means for performing processing for dividing the face image into a plurality of areas for the face image data;
Extraction means for performing the process of extracting the feature point data for each of the plurality of areas for all the areas that can be extracted;
Registration means for previously registering the extracted feature point data as registered feature point data of a registered face image with respect to a registered face image to be used as a reference for matching;
The matching feature point data of the matching target face image, which is the extracted feature point data of the matching target face image to be a matching target, and the registered feature point data are matched for each corresponding area, Based on the matching result, comprising a matching means for matching the matching target face image and the registered face image;
The collation means performs collation between the collation target face image and the registered face image based only on a collation result in an area where both the collation target feature point data and the corresponding registered feature point data are extracted. Do,
Person authentication device. The dividing unit performs processing for normalizing the matching target face image and dividing the face image into a plurality of areas.
The person authentication device according to claim 1. The dividing means performs processing for dividing the collation target face image into five areas of left eyebrow, right eyebrow, left eye, right eye, and nose / mouth / cheek.
The person authentication device according to claim 1 or 2. The dividing unit performs a process of dividing the collation target face image into seven areas of left eyebrow, right eyebrow, left eye, right eye, left cheek, right cheek, and nose / mouth.
The person authentication device according to claim 1 or 2. The dividing means performs a process of dividing the verification target face image into eight areas of a left eyebrow, a right eyebrow, a left eye, a right eye, a left cheek, a right cheek, a nose, and a mouth.
The person authentication device according to claim 1 or 2. As a result of the collation in the collating means, further comprising a notifying means for notifying only when the collation target face image and the registered face image match.
The person authentication device according to claim 1. By being installed and executed on a computer, the computer is caused to function as the person authentication device according to any one of claims 1 to 6.
Computer program.

Images