JP6429823B2 - Feature extraction device - Google Patents

Description

  The present invention relates to a feature amount extraction apparatus, and more particularly to a feature amount extraction apparatus that extracts a feature amount from an object image.

  2. Description of the Related Art Conventionally, there has been proposed a face image authentication device that authenticates a target person by collating a two-dimensional face image acquired by photographing the face of the target person with a registered face image. Such a face image authentication device calculates the degree of similarity by comparing the registered face image of the subject person acquired in advance with the input face image acquired at the time of use, and performs a match determination using the degree of similarity. Whether to authenticate is determined based on the result. For this reason, if there is a difference in facial orientation or facial expression between the registered face image and the input face image, the degree of similarity is lowered and the authentication error rate is increased.

  As a method for verifying the identity by collating the registered face image and the input face image, as disclosed in Non-Patent Document 1, from the luminance value sequence of each pixel in the face image whose size and position are normalized. A unique space (Eigenface) method for generating a feature vector subspace by principal component analysis, projecting the feature vector of the input face image and the registered face image onto the partial space, calculating similarity, and performing matching determination Is widely known.

  According to the prior art of Non-Patent Document 1, partial distortion occurs in a face image due to changes in face orientation and facial expression, and pixels corresponding to each element of a feature vector between a registered face image and an input face image When the positional relationship between the two changes, the degree of similarity becomes low, and there is a problem that whether or not authentication is possible cannot be correctly determined.

  To solve this problem, we estimate separately the subspace that represents the face image variation due to personality and the subspace that represents the face image variation due to variations in lighting conditions and facial expressions, and represents the face image variation due to individuality. Non-Patent Document 2 discloses a method for performing collation determination by projection onto a partial space.

  According to the prior art of Non-Patent Document 2, by projecting a feature vector onto a partial space representing the variation of a face image due to personality, the influence of variations in lighting conditions and facial expressions is reduced. Can be alleviated. In addition, unlike a space composed of eigenvectors having a large eigenvalue obtained by principal component analysis, a sharp feature remains in the subspace obtained by the prior art of Non-Patent Document 2, so even in a low-dimensional subspace, It is possible to perform collation reflecting common facial features such as eye and nose shapes.

M. Turk and A. Pentland, “Face Recognition using Eigenfaces”, IEEE Conference on Computer Vision and Pattern Recognition, Maui, HI, June 1991. Simon J.D.Prince, James H. Elder, “Probabilistic Linear Discriminant Analysis for Inferences About Identity”, IEEE 11th International Conference on Computer Vision, 2007

  However, in order to reflect minute features such as spots and moles in a subspace, it is necessary to configure the subspace with a very high number of dimensions, which is not realistic due to problems of calculation accuracy and processing time.

  An object of the present invention is to provide a feature quantity extraction apparatus that can extract fine features by simple processing.

  In order to achieve the above object, a feature quantity extraction apparatus according to the present invention is a feature quantity extraction apparatus that extracts a feature quantity from an object image obtained by imaging an object, and performs learning based on a plurality of learning object images. A reconstructed object that generates a reconstructed object image obtained by reconstructing the object image using the storage unit that stores a reconstructed basis that is a set of basis vectors obtained in advance and the object image and the reconstructed basis Image generation means, and feature quantity extraction means for extracting a feature quantity from a difference image between the object image and the reconstructed object image are configured.

  The object image according to the present invention is an input object image that has been input and a registered object image that has been registered, and the reconstructed object image generation means uses the input object image and the reconstructed image as the reconstructed object image. An input reconstructed object image is generated by reconstructing the input object image using a base, and the feature amount extraction unit is a difference image between the input object image and the input reconstructed object image as the feature amount. A first feature amount is extracted from a certain first difference image, a second feature amount is extracted from a second difference image that is a difference image between the registered object image and the input reconstructed object image, and the first feature amount is extracted. And the second feature quantity may further include a matching unit that performs image matching between the input object image and the registered object image.

  The object image according to the present invention is an input object image that has been input and a registered object image that has been registered, and the reconstructed object image generation means uses the input object image and the reconstructed image as the reconstructed object image. An input reconstructed object image obtained by reconstructing the input object image using a base, and a registered reconstructed object image obtained by reconstructing the registered object image using the registered object image and the reconstructed base, The feature amount extraction unit extracts, as the feature amount, a first feature amount from a first difference image that is a difference image between the input object image and the input reconstructed object image, and the registered object image and the registration A second feature amount is extracted from a second difference image that is a difference image from the reconstructed object image, and the input object image and the registered object image are obtained by comparing the first feature amount and the second feature amount. Matching It can further comprise means.

  The object image according to the present invention is an input object image that has been input and a registered object image that has been registered, and the reconstructed object image generation unit includes the registered object image and the reconstructed image as the reconstructed object image. A registered reconstructed object image is generated by reconstructing the registered object image using a base, and the feature amount extraction unit is a difference image between the input object image and the registered reconstructed object image as the feature amount. A first feature amount is extracted from a certain first difference image, a second feature amount is extracted from a second difference image that is a difference image between the registered object image and the registered reconstructed object image, and the first feature amount is extracted. And the second feature quantity may further include a matching unit that performs image matching between the input object image and the registered object image.

  The feature amount extraction unit according to the present invention extracts the first feature amount from the first difference image and the input object image, and extracts the second feature amount from the second difference image and the registered object image. can do.

  The feature amount extraction means according to the present invention can extract a feature amount from the image obtained by dividing each pixel of the first difference image by the variance value of the reconstruction residual obtained by the learning as the first feature amount. .

  The feature amount extraction unit according to the present invention extracts a pixel position having a pixel value equal to or greater than a predetermined threshold in the first difference image as the first feature amount, and a pixel having a pixel value equal to or greater than the predetermined threshold in the second difference image. The position can be extracted as the second feature amount.

  The input object image according to the present invention is a plurality of input object images captured at different times, and the reconstructed object image generation means generates the input reconstructed object image for each of the plurality of input object images, The feature amount extraction unit obtains the first difference image between each input object image and the input reconstructed object image corresponding to the input object image, and based on the first difference image obtained for each input object image The first feature amount can be extracted from the obtained integrated difference image.

  The number of basis vectors included in the reconstruction basis according to the present invention is determined in advance as a residual between the learning object image and the reconstruction object image reconstructed using the learning object image and the reconstruction basis. It can be set to be within the specified range.

  As described above, according to the feature quantity extraction device of the present invention, a reconstructed object image obtained by reconstructing an object image using the object image and the reconstructed base is generated, and the object image and the reconstructed object image are By extracting the feature amount from the difference image, it is possible to extract a fine feature by a simple process.

It is the schematic which shows the structure of the entrance / exit management system which concerns on embodiment of this invention. It is a block diagram which shows the structure of the face image authentication apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the feature point and rectangle of a face image. It is a figure which shows an example of the face area image cut out by the rectangle. (A) is a figure which shows the example of a face area image, (B) is a figure which shows the example of a reconstruction face image. It is a figure which shows the example of a difference image. It is a figure which shows the result of having performed the threshold value process with respect to a difference image. It is a flowchart which shows the operation | movement of the authentication process by the face image authentication apparatus which concerns on embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the present embodiment, an example will be described in which the present invention is applied to an access control system, an input face image and a registered face image are used as object images, and the input face image and the registered face image are collated.

<System configuration>
Hereinafter, referring to FIG. 1 schematically showing an installation example of the entrance / exit management system 100 for managing entrance / exit of the room entrance and FIG. 2 showing a schematic configuration of the entrance / exit management system 100 to which the present invention is applied, The configuration of the embodiment of the present invention will be described.

(Access control system 100)
The access management system 100 includes an imaging device 1, an electric lock control device 2, and a face image authentication device 3. In the schematic diagram shown in FIG. 1, the imaging device 1, the electric lock control device 2, and the face image authentication device 3 are arranged in the vicinity of the entrance / exit of the room. As shown in FIG. 1, the imaging apparatus 1 has a shooting direction directed slightly downward on a wall surface or ceiling near the entrance 101 so that the passage leading to the entrance 101 can be taken as a shooting area, and toward the passage side leading to the entrance 101. It is attached in the state of facing. Thereby, the imaging device 1 can image the person 110 heading toward the entrance 101 along the traveling direction 102.

  The face image authentication device 3 manages the entrance / exit of the entrance 101 of the room by controlling the electric lock 104 via the electric lock control device 2. The entrance 101 is provided with a door 103, and the door 103 is provided with an electric lock 104. The electric lock 104 can be locked and unlocked by the face image authentication device 3. The electric lock 104 always locks the door 103, and is unlocked only for a fixed time (for example, 5 seconds) if it can be authenticated that the face image authentication device 3 is permitted to pass. The electric lock 104 is automatically locked when the door is closed and the door is closed and a predetermined time elapses after the door is opened.

  Hereinafter, the face image authentication device 3 will be described in detail with reference to FIG. FIG. 2 is a diagram showing a schematic configuration of the access control system 100 to which the present invention is applied.

(Imaging device 1)
The imaging device 1 is a surveillance camera that captures a predetermined imaging region as shown in FIG. The imaging device 1 is connected to the face image authentication device 3 and outputs the acquired input image to the face image authentication device 3.

(Electric lock control device 2)
The electric lock control device 2 is connected to the face image authentication device 3, and locks or unlocks the electric lock 104 according to a signal from the face image authentication device 3.

(Face image authentication device 3)
The face image authentication device 3 includes a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), peripheral circuits, terminals, and various memories. The face image authentication device 3 authenticates a person's face image captured by the imaging device 1, and If it is authenticated that the person is permitted to pass 101, the control signal of the electric lock 104 is output to the electric lock control device 2. The face image authentication device 3 is functionally composed of a face image acquisition unit 10, an output unit 20, a storage unit 30, and an image processing unit 40. Hereinafter, each part of the face image authentication device 3 will be described in detail.

(Face image acquisition unit 10)
The face image acquisition unit 10 is connected to the imaging device 1, extracts a face area image including a human face from an input image captured by the imaging device 1, and outputs it to the image processing unit 40. When a plurality of human faces are photographed in the input image, face area images are sequentially extracted for each of the plurality of persons and output to the image processing unit 40.

  When extracting the face area image, the cut-out position and size of the face area image are normalized so that an image reconstruction process described later can be performed. Specifically, four feature points of the center of both eyes, the nose vertex, and the center of the mouth are extracted from the face area image, and the face image is enlarged and reduced so that the number of pixels between both eyes becomes a predetermined value. Cut out a rectangle of a predetermined size centered on the barycentric positions of the four feature points. The number of pixels between eyes and the cutout size of the rectangle here are the same values as the number of pixels between eyes of the learning face image and the cutout size of the rectangle that are aligned prior to the calculation of the reconstruction basis.

  Many face area image extraction methods and face feature point extraction methods have been proposed in the past, and publicly known methods may be employed as appropriate. For example, a method of extracting a face image or an image around a facial feature point with a learned filter called a discriminator, or a binarized edge image of an input image is generated, and an elliptical portion that is a face shape in the edge image Alternatively, a method of detecting a characteristic shape around the face feature point may be employed.

(Output unit 20)
The output unit 20 is an interface connected to the electric lock control device 2 that is an external connection device and a control circuit thereof. When the output unit 20 receives a signal indicating that the person has been successfully authenticated from the image processing unit 40, the output unit 20 outputs a signal indicating that the authentication has been successful to the connected electric lock control device 2.

(Storage unit 30)
The storage unit 30 includes a semiconductor memory such as a ROM (Read Only Memory) and a RAM (Random Access Memory), or a magnetic recording medium and its access device or an optical recording medium and its access device. The storage unit 30 stores a computer program and various data for controlling the face image authentication device 3, and inputs / outputs such information to / from the image processing unit 40. The various data includes data indicating the reconstructed base 31 and the registered face image 33 of the registrant permitted to pass through the entrance 101.

(Reconstruction base 31)
The reconstruction base 31 is a set of basis vectors representing a partial space for reconstructing a human face image. The number of elements of the basis vector is the same as the number of pixels of the image output by the reconstruction unit 41, and the human face image is reconstructed by a linear sum of the basis vectors. The number of basis vectors included in the reconstructed basis 31 is finite. For example, a predetermined number of basis vectors can be obtained by applying factor analysis to a large number of human face images.

Hereinafter, a procedure for applying factor analysis to N face images prepared for learning and obtaining M basis vectors will be described in detail. In the following description, the number of pixels of the image output by the reconstruction unit 41 is D, and the face image of a person to which factor analysis is applied is a D-dimensional vector.

, The average of N face images is also a D-dimensional vector

, D × M matrix of subspace consisting of M basis vectors

Represented by

  Prior to learning, the face images of N persons are aligned. As an alignment method, for example, there is a method of aligning the number of pixels between both eyes and the center of gravity of the face image. Specifically, as shown in FIG. 3, feature points 301 at the center of both eyes, the nose apex, and the center of the mouth are extracted from the face image 300, and the face image is scaled so that the number of pixels between the eyes becomes a predetermined value. At the same time, a rectangle 303 having a predetermined size centered on the barycentric position 302 of the four feature points is cut out. FIG. 4 shows an example of a face image that is enlarged and reduced so that the number of pixels between both eyes is 100 and cut out with a rectangle of 192 pixels per side centered on the feature point centroid.

Human face image

Is an M-dimensional vector representing coordinates in the subspace

And a D-dimensional vector representing the residual of the reconstruction

Is assumed to be expressed by the equation (1).

(1)

A D-dimensional vector representing the residual of the reconstruction

Is mean 0, variance matrix

According to the normal distribution of

Maximize the likelihood of

The EM (expectation-maximization) algorithm for obtaining is given by equation (2).

(2)

First,

Set an arbitrary initial value to, and use the E-Step shown in Equation (2)

and

Expected value

Ask for. Next, it was calculated by E-Step by M-Step
,
New with

Ask for. By repeating this E-Step and M-Step, the person's face image for learning

Maximize the likelihood for

Is obtained. Obtained in this way

Is stored in the storage unit 30 as the reconstruction base 31.

Reconstruction base

Is smaller than the dimension D of the face image, and the learning face image is arbitrary.

And face image

Reconstruct basis

Reconstructed face image reconstructed with

Is set to such a number that the residual E _i becomes sufficiently small. The residual E _i is expressed by a difference amount in units of pixels expressed by the expression (3), and a face image for arbitrary learning.

The number of bases M is set so that the residual E _i falls within a predetermined range (for example, a range from 10 to 20).

(3)

Reconstructed face image

Is the face image obtained by the above equation (2)

Reconstruction base corresponding to

Coordinates in the subspace consisting of

Expected value

Can be obtained by the equation (4).

(4)

  In addition to the factor analysis method described above, various known multivariate analysis methods such as principal component analysis, independent component analysis, and sparse coding can be applied to the calculation method of the reconstruction basis 31. Even when the multivariate analysis method is applied, the number M of bases constituting the reconstructed base is sufficiently smaller than the dimension D of the face image, and the reconstructed face image by the face image and the reconstructed base 31 is used. And set the number so that the residual is sufficiently small.

(Registered face image 33)
The registered face image 33 is a face image of a person permitted to enter and exit in advance, and is used for authentication with the input face image. One or more face images are stored for each person.

(Image processing unit 40)
The image processing unit 40 includes a so-called computer such as a memory, a peripheral circuit thereof, and a microprocessor thereof. The image processing unit 40 performs various processes on the face area image acquired from the face image acquisition unit 10 while referring to the storage unit 30. The processing result is output to the output unit 20. The image processing unit 40 includes a reconstruction unit 41, a feature amount calculation unit 42, and a collation unit 43. Each means of the image processing unit 40 is a functional module realized by software operating on a microprocessor.

(Reconstruction means 41)
The reconstruction unit 41 reconstructs the face area image whose face position and size have been normalized by the face image acquisition unit 10 with reference to the reconstruction base 31 stored in the storage unit 30.

Hereinafter, the vector of the face area image output by the face image acquisition unit 10 is

The image reconstruction process in the reconstruction unit 41 will be described in detail.

Is assumed to be expressed in the form of Equation (5), similar to the learning face image of the reconstruction basis,

Expected value
Is obtained by equation (5), similar to the equation shown by E-Step in equation (2).

(5)

(6)

Sought here

Expected value

And the input reconstructed face image according to equation (7)

Get.

(7)

In the method described above, in determining the contribution of the basis used for reconstruction, the face image of the person for learning in each basis direction

Is not considered. That is, the reconstructed face image obtained by the above-described method is different from the face image obtained by projecting on the space composed of eigenvectors having a large eigenvalue obtained by principal component analysis, and the face such as the shape of eyes and nose There is a feature that makes a sharp image reflecting the feature. FIG. 5 shows an example of a face area image 500 and an example of a reconstructed face image 501 by the method described above.

  The reconstruction unit 41 reconstructs the registered face image 33 with reference to the reconstruction base 31 stored in the storage unit 30 in the same manner as the face area image acquired by the face image acquisition unit 10, and performs registration reconfiguration. A constituent face image is obtained.

(Feature amount calculation means 42)
The feature amount calculating means 42 is a face area image output by the face image acquisition unit 10.

And the input reconstructed face image obtained by the reconstructing means 41

The first difference image that is the difference image is obtained, and the face matching feature amount used for matching the face image is calculated from the first difference image.

Input face image

If the brightness value is 8bit representation of 0-255, usually the reconstructed face image

Similarly, the luminance value takes a value in the range of 0 to 255. Therefore, in order to apply the conventional method for calculating the feature amount, in the case of the 8-bit expression with the luminance value of 0 to 255 for the difference image, for example, by applying a range conversion as shown in Equation (8), Difference image

Get.

(8)

As described above, the reconstructed face image obtained by the reconstructing means 41

5B has a feature that becomes a sharp image reflecting facial features such as the shape of eyes and nose as shown in the reconstructed face image 501 in FIG.

This reflects the fine features such as moles and spots that differ in the position of occurrence among individuals. Figure 6 shows the difference image

, And a mole feature 601 appearing on the difference image 600.

The feature quantity calculation means 42 is at least the first difference image obtained by the above procedure.

To calculate a first face matching feature value used for face image matching. Since various known methods such as LBP (Local Binary Pattern) can be used for calculating the face matching feature value, detailed description thereof is omitted. The feature quantity calculation means 42 is a first difference image.

Not only face image

And input reconstruction face image

Also, the face matching feature value may be extracted.

  Similar to the face area image output by the face image acquisition unit 10, the feature amount calculating unit 42 also calculates a difference image between the registered face image 33 and the registered reconstructed face image obtained by the reconstructing unit 41 for the registered face image 33. A second difference image is obtained, and a second face matching feature amount used for face image matching is calculated from the second difference image.

  In the present embodiment, the image processing unit 40 obtains a registered reconstructed face image and a second difference image from the registered face image 33 for all the registered face images 33 stored in the storage unit 30, and preliminarily stores the second face collation. The feature amount is calculated in advance. Note that the image processing unit 40 may calculate the second face matching feature amount for each registered face image 33 each time the face area image is collated.

(Verification means 43)
The collation unit 43 collates the face area image output from the face image acquisition unit 10 with all the registered face images 33 stored in the storage unit 30.

  Hereinafter, the collation process in the collation means 43 will be described in detail. First, the similarity between the first face matching feature value calculated from the face area image and the second face matching feature value calculated from the registered face image 33 is obtained, and the registered face image 33 that maximizes the similarity is selected. When the similarity is equal to or greater than the authentication threshold, it is determined that the face shown in the face area image is the face of the registrant corresponding to the registered face image 33.

  For the method of calculating the similarity between the first face matching feature value calculated from the face area image and the second face matching feature value calculated from the registered face image 33, a feature value such as a normalized correlation value or the reciprocal of the L2 distance is used. Since various known methods may be selected in accordance with the feature amount calculated by the calculation means 42, detailed description thereof is omitted.

  If the collation means 43 determines that the face is the registrant's face, it outputs a signal indicating successful authentication to the output unit 20 and causes the electric lock control device 2 to output a signal for performing unlock control.

<Operation of face image authentication device>
The operation of the authentication process by the face image authentication apparatus 3 to which the present invention is applied will be described below with reference to the flowchart shown in FIG. The operation described below is executed every time one input image is acquired.

  First, the face image acquisition unit 10 of the face image authentication device 3 acquires an input image captured by the imaging device 1 (step S101). Then, the face image acquisition unit 10 extracts a face area image from the input face image (step S102). The face image acquisition unit 10 determines whether or not one or more face area images have been extracted (step S103). If no face area image has been extracted, the following process is not performed and the authentication process is performed. finish. On the other hand, when one or more face area images are extracted, the face image acquisition unit 10 shifts the processing to step S104.

  The following steps S104 to S109 are performed for each face area image extracted by the face image acquisition unit 10.

The reconstruction unit 41 refers to the reconstruction base 31 stored in the storage unit 30 and refers to the face area image.

Reconstructed face image input from

Is obtained (step S104). Next, the feature quantity calculating means 42 is a face area image.

And the input reconstructed face image output by the reconstruction means 41

First difference image from

(Step S105), the first difference image

The first face matching feature amount is obtained from (step S106).

  Next, the matching unit 43 calculates the similarity between the first face matching feature value calculated by the feature value calculating unit 42 and the second face matching feature value of all the registered face images 33 stored in the storage unit 30. Then, the registered face image 33 having the highest similarity is selected (step S107). When the similarity in the selected registered face image 33 is equal to or higher than the authentication threshold (step S108), it is determined that the face of the person shown in the face area image is the face of the registrant corresponding to the registered face image 33; A signal indicating successful authentication is output to the output unit 20 (step S109). When a signal indicating successful authentication is output to the output unit 20, a signal for performing unlock control on the electric lock control device 2 is output from the output unit 20. On the other hand, when the similarity in the selected registered face image 33 is less than the authentication threshold, the matching unit 43 does not perform any particular processing.

  When the processes of steps S104 to S109 are completed for all face area images, the image processing unit 40 ends a series of steps.

  As described above, the face image authentication apparatus according to the embodiment of the present invention generates an input reconstructed face image obtained by reconstructing a face area image using the face area image and the reconstruction base, and By extracting the feature amount from the difference image between the image and the input reconstructed face image, it is possible to extract a face matching feature amount that represents a fine feature unique to an individual such as a mole or a stain from the face image. As a result, although the shapes of organs such as eyes and nose and mouth such as twins are very similar, it is possible to obtain high identification accuracy in identifying persons with different individual features such as moles and spots.

<Modification>
The preferred embodiments of the present invention have been described above, but the present invention is not limited to these embodiments. For example, in the present embodiment, the face image acquisition unit 10 processes the input image captured by the imaging device 1 and outputs the processed image to the image processing unit 40. However, the face image acquisition unit 10 acquires the input image from a medium such as a hard disk, May be extracted and output to the image processing unit 40.

  In this embodiment, since the authentication result is used to pass or not, the authentication result is output to the electric lock control device 2. However, the authentication result is transmitted via a communication line such as a general public line or a mobile phone line. May be output to an external device such as a monitoring center device. In addition, a lamp for displaying the authentication result may be provided in the face image authentication device 3 without outputting it to the external device, and the authentication result may be notified to the passerby.

  In the above embodiment, the input reconstructed face image is used as the reconstructed face image when obtaining the first difference image, and the registered reconstructed face image is used as the reconstructed face image when obtaining the second difference image. However, as the reconstructed face image, each of the first difference image and the second difference image may be obtained using the input reconstructed face image without using the registered reconstructed face image. As the face image, each of the first difference image and the second difference image may be obtained using the registered reconstructed face image without using the input reconstructed face image.

  When obtaining each of the first difference image and the second difference image using the input reconstruction face image as the reconstruction face image, the difference image between the face area image and the input reconstruction face image is obtained as the first difference image, A first face matching feature is extracted from the obtained first difference image, a difference image between the registered face image and the input reconstructed face image is obtained as a second difference image, and the second face matching feature is obtained from the obtained second difference image. Extract the amount.

  In addition, when obtaining each of the first difference image and the second difference image using the registered reconstructed face image as the reconstructed face image, the difference image between the face area image and the registered reconstructed face image is used as the first difference image. The first face matching feature value is extracted from the obtained first difference image, the difference image between the registered face image and the registered reconstructed face image is obtained as the second difference image, and the second face is obtained from the obtained second difference image. Extract matching features.

Further, in the present embodiment, the feature amount calculating unit 42 calculates the difference image vector by a simple difference calculation.

However, as shown in FIG. 6, a difference also appears in a portion where the reconstruction residual is large. So, input face image vector

And reconstruction face image vector

For each pixel, the variance matrix of the reconstruction residual obtained when calculating the reconstruction base 31

The influence of the residual may be reduced by obtaining the first difference image, for example, by dividing by the diagonal component. For example, the difference image as shown in Expression (9)

The pixel value of each pixel may be obtained.

(9)
However,

Respectively

Elements of

It is.

Further, the difference image obtained by the feature amount calculating means 42

A pixel having an absolute value greater than or equal to a predetermined threshold value may be extracted, and position information of the pixel may be extracted as a face matching feature amount representing position information having individual-specific features. In this case, a pixel having an absolute value greater than or equal to a predetermined threshold is extracted from the first difference image for the face area image, and the position information of the pixel is extracted as the first face matching feature amount. A pixel having an absolute value equal to or greater than a predetermined threshold is extracted from each of the second difference images, and position information of the pixel is extracted as a second face matching feature amount.

Moreover, as shown in FIG. 7, the difference image calculated | required by said Formula (9).

For each of the pixels, threshold processing for setting the pixel value to 0 may be performed when the pixel value is less than a predetermined threshold. Image features and peak positions may be extracted from these images as face matching feature values and compared. Thereby, collation of the positional information which has the characteristic peculiar to an individual is attained.

Further, the difference image obtained by the feature amount calculating unit 42 is accumulated in the time direction, and the averaged integrated difference image is converted into the difference image.

It is also good. In this case, the reconstruction unit 41 obtains an input reconstruction face image for each of a plurality of input face images captured at different times, and the feature amount calculation unit 42 corresponds to the input face image for each input face image. What is necessary is just to extract the 1st face collation feature-value from the integrated difference image obtained by calculating | requiring the 1st difference image with the input reconstruction face image to perform, and averaging the 1st difference image calculated | required about each input face image .

  Moreover, although the case where this invention is applied to a face image authentication apparatus was demonstrated to the example, it is not limited to this. For example, the present invention may be applied to a device that estimates the state of the face represented by the input face image. In this case, the state of the face may be estimated from the feature amount extracted for the input face image.

  In addition, the case where image matching is performed by extracting the feature amount of the input face image has been described as an example. However, the present invention is not limited to this, and the feature amount of an image representing an object having a shape other than a face is extracted. Then, image collation may be performed. For example, a feature amount of an image representing a car may be extracted and image collation may be performed.

  As described above, those skilled in the art can make various changes in accordance with the embodiment to be implemented within the scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Imaging device 2 Electric lock control device 3 Face image authentication apparatus 10 Face image acquisition part 30 Storage part 31 Reconstruction base 33 Registered face image 40 Image processing part 41 Reconstruction means 42 Feature amount calculation means 43 Collation means 100 Access control system 300 Face image 501 Reconstructed face image 600 Difference image

Claims (8)

A feature amount extraction device that extracts a feature amount from an object image obtained by imaging an object,
The object image is an input object image that has been input and a registered object image that has been registered,
A storage unit that stores in advance a reconstructed basis that is a set of basis vectors obtained by performing learning based on a plurality of learning object images;
Reconstructed object image generation means for generating an input reconstructed object image obtained by reconstructing the input object image using the input object image and the reconstructed base;
A first difference image that is a difference image between the input object image that extracts a first feature amount from a first difference image that is a difference image between the input object image and the input reconstructed object image and the input reconstructed object image It extracts the first feature quantity from the feature quantity extraction means for extracting a second feature value from a second differential image wherein a differential image between the registered object image and the input reconstructed image,
Collating means for collating the input object image and the registered object image by comparing the first feature quantity and the second feature quantity;
A feature amount extraction device. A feature amount extraction device that extracts a feature amount from an object image obtained by imaging an object,
The object image is an input object image that has been input and a registered object image that has been registered,
A storage unit that stores in advance a reconstructed basis that is a set of basis vectors obtained by performing learning based on a plurality of learning object images;
Registration reconstituted the registration object image using reconstructed input reconstructed image, and said reconstituted basement and the registered object images the input object image using said reconstituted basement and the input object image Reconstructed object image generating means for generating a reconstructed object image;
The first feature image is extracted from the first difference image that is a difference image between the input object image and the input reconstructed object image and the input reconstructed object image, and the registered object image and the registered reconstructed object image A feature amount extraction means for extracting a second feature amount from the second difference image and the registered reconstructed object image ,
Collating means for collating the input object image and the registered object image by comparing the first feature quantity and the second feature quantity;
A feature amount extraction device. A feature amount extraction device that extracts a feature amount from an object image obtained by imaging an object,
The object image is an input object image that has been input and a registered object image that has been registered,
A storage unit that stores in advance a reconstructed basis that is a set of basis vectors obtained by performing learning based on a plurality of learning object images;
Reconstructed object image generating means for generating a registered reconstructed object image obtained by reconstructing the registered object image using the registered object image and the reconstructed base;
A first feature amount is extracted from a first difference image that is a difference image between the input object image and the registered reconstructed object image, and a second image that is a difference image between the registered object image and the registered reconstructed object image. Feature amount extraction means for extracting a second feature amount from the difference image ;
Collating means for collating the input object image and the registered object image by comparing the first feature quantity and the second feature quantity;
A feature amount extraction device. The feature amount extraction unit extracts the first feature amount from the first difference image and the input object image, and extracts the second feature amount from the second difference image and the registered object image. The feature amount extraction apparatus according to claim 1 or 3 . The feature extraction means, said as a first feature amount, claims 1 to extract a feature from the image divided by the dispersion value of the reconstructed residual obtained by the learning of each pixel of the first difference image The feature quantity extraction device according to any one of claims 4 to 4 . The input object images are a plurality of input object images taken at different times,
The reconstructed object image generating means generates the input reconstructed object image for each of the plurality of input object images,
The feature amount extraction unit obtains the first difference image between each input object image and the input reconstructed object image corresponding to the input object image, and based on the first difference image obtained for each input object image feature extraction apparatus according to claim 1 or claim 2 from the resulting integrated difference image to extract the first feature amount. The feature amount extraction unit extracts, as the first feature amount, a pixel position having a pixel value greater than or equal to a predetermined threshold in the first difference image, and determines a pixel position having a pixel value greater than or equal to the predetermined threshold in the second difference image. feature extraction apparatus according to any one of claims 1 to 6 to extract as the second feature amount. The number of basis vectors included in the set of basis vectors is determined in advance as a residual between the learning object image and the reconstructed object image reconstructed using the learning object image and the reconstructed basis. The feature quantity extraction device according to any one of claims 1 to 7 , which is set to be within a range.