JP6355306B2 - Face matching device, face matching method, and face matching program - Google Patents

Description

  According to the present invention, registration data including a face image of a person registered in advance or a feature amount of the face image, a face image of a person to be collated (hereinafter referred to as “verification target person”), or the face image of the face image The present invention relates to a face collation apparatus, a face collation method, and a face collation program that collate input data composed of feature amounts.

  Conventionally, in order to determine whether or not the person to be collated matches a person registered in advance, the registered data including the face image of the person registered in advance or the feature amount of the face image, and the person to be collated 2. Description of the Related Art A face matching technique that performs a matching process with a face image or input data composed of a feature amount of the face image is known.

  For example, in Patent Document 1, a feature amount for each small region obtained by dividing input data is extracted, a similarity with a feature amount for each small region obtained by dividing registered data is obtained, and the input data and the registered data are collated. Techniques for processing are disclosed. In Patent Document 2, distance values between a plurality of partial areas in input data and corresponding partial areas in registration data are calculated, and the input data and registration data are determined by the integrated distance values calculated in ascending order of the distance values. A technique for matching processing is disclosed.

  As described above, according to the conventional face matching technology, the face matching process is performed using the similarity or distance value between the input data and the registered data. May decrease. For this reason, a conventional technique for performing face matching processing while reducing the influence of such illumination is known.For example, in consideration of the point where the illumination is irradiated from a specific direction, the pixel value of the portion affected by the illumination is determined. An image collating apparatus for correction is disclosed (see Patent Document 3).

Japanese Patent No. 4175413 Japanese Patent No. 4161659 JP 2002-24830 A

  However, the method of Patent Document 3 can be applied only when illumination is always emitted from a specific direction, and when it is not possible to specify which part of the face image is affected by illumination, Therefore, there is a problem that it is not possible to exclude the influence of illumination dynamically and adaptively.

  For example, if a mobile lighting device is installed near the face matching device, or if the installation location of the face matching device is changed, the effect of lighting will be affected if the pixel to be corrected is not changed each time. It cannot be removed.

  For this reason, how to dynamically and adaptively reduce the influence of lighting on the face matching process is an important issue. Such a problem is not limited to the influence of the illumination on the face collation apparatus, but is similarly caused by the influence of various light sources such as sunlight on the face collation apparatus. For example, when the face matching device is installed in an indoor area or outdoors where sunlight enters depending on the time of day, the irradiation direction of sunlight is not constant. This is because it cannot be removed well. In addition, in this specification, the lighting environment including the lighting condition by artificial lighting and the lighting condition by natural light such as sunlight is referred to as an illumination environment.

  The present invention has been made to solve the above-described problems of the prior art, and is a face matching device and face matching capable of dynamically and adaptively reducing the influence on face matching by a light source such as sunlight and lighting. It is an object to provide a method and a face matching program.

In order to solve the above-described problems and achieve the object, the present invention performs a collation process on input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a previously registered face image or its feature quantity. Illumination influence reference data calculation that is a face collation device and calculates illumination influence reference data used as a reference when calculating the influence degree of a light source in a face image using learning data including a plurality of face images with different illumination environments An influence degree calculating means for calculating an influence degree of a light source in each local region of the input data and / or the registration data based on the illumination influence reference data calculated by the illumination influence reference data calculating means; Similarity calculating means for calculating the similarity between the feature amount in each local region of the input data and the feature amount in the corresponding local region of the registered data; Collation processing means for collating the input data with the registered data based on the similarity between the feature amounts in each local area calculated by the similarity calculation means and the influence degree calculated by the influence degree calculation means The lighting influence reference data calculation means includes, from the learning data including a plurality of face images with different lighting environments, lighting environment reference data serving as a reference when obtaining lighting parameters corresponding to the lighting environment; and wherein the calculation child a characteristic variation reference data serving as a reference for obtaining the characteristic variation occurring with the feature value in the local region under illumination environment corresponding to the illumination parameters as the lighting effect reference data.

Further, in the above invention, before Symbol influence calculation means, said calculating illumination parameters of the input data and / or the registration data using the illumination environment reference data, the feature change and the calculated illumination parameters A feature weight calculating means for calculating a feature variation rate in each local region of the input data and / or the registered data using reference data, and calculating a weight of each local region according to the feature variation rate; The collation processing unit performs a collation process on the input data and the registration data by weighting the similarity of each local region using the region weight calculated by the region weight calculation unit.

  Further, the present invention is the above invention, wherein the illumination influence reference data calculation means performs a resolution reduction process for reducing resolution on the learning data including a plurality of face images having different illumination environments, and the low resolution The lighting environment reference data is calculated from the learning data subjected to the conversion process, and the influence calculation unit performs a resolution reduction process for reducing the resolution of the input data and / or the registration data, and The illumination parameter is calculated using the input data and / or the registration data subjected to resolution processing and the illumination environment reference data.

  Further, in the present invention according to the above invention, the illumination influence reference data calculation unit includes the number of pixels of the learning data subjected to the resolution reduction processing as the number of rows or columns, and the number of the plurality of face image data as columns. A pixel value matrix having a number or a row number is generated, an eigenvector of the pixel value matrix is calculated as the illumination environment reference data, and the influence calculation unit includes the input data and / or the resolution reduction process The product of each pixel value of the registration data and the illumination environment reference data is calculated as the illumination parameter.

  Further, the present invention is the above invention, wherein the illumination influence reference data calculation means calculates a product of each pixel value of the face image of the learning data subjected to the resolution reduction processing and the illumination environment reference data as the illumination parameter. And calculating the feature variation amount of each local region from the learning data before the resolution reduction processing, and assuming that the feature variation amount of the local region is a linear transformation of the illumination parameter for each local region The linear transformation coefficient matrix is calculated as the feature variation reference data, and the influence calculation means includes the illumination parameter calculated from the input data and / or the registration data subjected to the resolution reduction processing, and A product with the feature variation reference data is calculated as a feature variation amount in the input data and / or the registered data.

  Further, the present invention is the above invention, wherein the area weight calculation unit is configured such that when the influence degree calculation unit calculates a feature variation rate in each local area for both the input data and the registered data, The region weight is calculated using the larger feature variation rate.

The present invention is also a face collation method for collating input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a face image or its feature quantity registered in advance, and having a different illumination environment By using the learning data including a plurality of face images, the lighting influence reference data calculating step for calculating the lighting influence reference data serving as a reference when calculating the influence degree of the light source in the face image, and the lighting influence reference data calculating step Based on the calculated illumination influence reference data, an influence degree calculating step for calculating an influence degree of a light source in each local area of the input data and / or the registration data, a feature amount in each local area of the input data, and the A similarity calculation step of calculating a similarity between the feature values in the corresponding local region of the registered data, and the similarity calculation step A similarity of the feature amount in each local region calculated Ri, based on the degree of influence calculated by the influence degree calculating step, viewed contains a collation processing step of collating processing the input data the registration data , In the lighting influence reference data calculation step, the lighting environment reference data serving as a reference when obtaining the lighting parameters corresponding to the lighting environment from the learning data including a plurality of face images having different lighting environments; and the lighting parameter the feature variation reference data serving as a reference for obtaining the characteristic variation occurring with the feature value in the local region under illumination environment corresponding features that you calculated as the lighting effect reference data.

The present invention is also a face collation program for collating input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a face image or its feature quantity registered in advance, and having a different illumination environment By using the learning data including a plurality of face images, the lighting influence reference data calculation procedure for calculating the lighting influence reference data serving as a reference when calculating the influence degree of the light source in the face image, and the illumination influence reference data calculation procedure Based on the calculated illumination influence reference data, an influence calculation procedure for calculating an influence degree of a light source in each local area of the input data and / or the registration data, a feature amount in each local area of the input data, and the The similarity calculation procedure for calculating the similarity between the feature values in the corresponding local region of the registered data, and the similarity calculation procedure A similarity of the feature amount in a local region, based on the degree of influence calculated by the influence degree calculation procedure, to execute a collation process procedure of collation processing and the input data and the registered data to the computer, the illumination In the influence reference data calculation procedure, the illumination environment reference data serving as a reference for obtaining the illumination parameter corresponding to the illumination environment from the learning data including a plurality of face images having different illumination environments, and the illumination parameter the feature variation reference data serving as a reference for obtaining the characteristic variation occurring with the feature value in the local region under illumination environment characterized that you calculated as the lighting effect reference data.
Further, the present invention is a face matching device that performs matching processing of input data consisting of a face image of a person to be matched or a feature amount thereof with registered data consisting of a face image registered in advance or the feature amount thereof, Each local area of the input data and / or the registration data is calculated based on illumination influence reference data, which is a reference for calculating the influence degree of the light source in the face image, calculated using learning data including a plurality of different face images. Influence degree calculation means for calculating the influence degree of the light source in the area, and similarity calculation means for calculating the similarity degree between the feature quantity in each local area of the input data and the feature quantity in the corresponding local area of the registered data Based on the similarity of the feature amount in each local area calculated by the similarity calculation means and the influence calculated by the influence calculation means. Collation processing means for collating input data with the registered data, and the illumination influence reference data includes illumination environment reference data serving as a reference for obtaining an illumination parameter corresponding to the illumination environment, and the illumination parameter. And feature variation reference data serving as a reference for obtaining a feature variation amount generated in the feature amount in the local region under a corresponding illumination environment.
The present invention is also a face matching program for matching input data consisting of a face image of a person to be matched or a feature value thereof with registered data consisting of a face image registered in advance or a feature value thereof. Each local area of the input data and / or the registration data is calculated based on illumination influence reference data, which is a reference for calculating the influence degree of the light source in the face image, calculated using learning data including a plurality of different face images. A degree-of-influence calculation procedure for calculating a degree of influence of a light source in a region, and a degree-of-similarity calculation procedure for calculating a similarity between a feature amount in each local region of the input data and a feature amount in a corresponding local region of the registered data And the similarity of the feature amount in each local area calculated by the similarity calculation procedure and the influence calculated by the influence calculation procedure , Causing the computer to execute a collation processing procedure for collating the input data with the registered data, and the illumination influence reference data includes illumination environment reference data serving as a reference when obtaining an illumination parameter corresponding to the illumination environment And feature variation reference data serving as a reference for obtaining a feature variation amount generated in the feature amount in the local region under an illumination environment corresponding to the illumination parameter.

According to the present invention, it is possible to reduce the dynamic and adaptive affected by the light source.

FIG. 1 is an explanatory diagram for explaining the concept of face matching processing by the face matching device. FIG. 2 is a block diagram showing the configuration of the face matching device. FIG. 3 is an explanatory diagram for explaining generation of illumination influence reference data. FIG. 4 is an explanatory diagram for explaining the feature amount distribution and the feature variation rate. FIG. 5 is an explanatory diagram for describing details of the illumination influence reference data generation unit illustrated in FIG. 2. FIG. 6 is an explanatory diagram for explaining generation of a feature amount and a feature variation rate. FIG. 7 is an explanatory diagram for explaining details of registration and collation. FIG. 8 is an explanatory diagram for explaining a case where the logistic sigmoid function is used for calculating the region weight. FIG. 9 is a flowchart illustrating a processing procedure related to learning and registration of the face matching device. FIG. 10 is a flowchart showing the processing procedure of the illumination influence reference data generation processing. FIG. 11 is a flowchart illustrating the processing procedure of the feature variation rate calculation processing. FIG. 12 is a flowchart showing a processing procedure related to the collation of the face collation apparatus. FIG. 13 is an explanatory diagram for describing imaging of learning data. FIG. 14 is an explanatory diagram when each functional unit of the face collation apparatus is realized by a program.

  Exemplary embodiments of a face matching device, a face matching method, and a face matching program according to the present invention will be described below in detail with reference to the accompanying drawings. In the embodiment described below, the case of removing the influence of illumination will be described, but the present invention can be similarly applied to the case of removing the influence of other light sources such as sunlight.

  First, the concept of the face matching process of the face matching device 10 according to the present embodiment will be described. FIG. 1 is an explanatory diagram for explaining the concept of face collation processing by the face collation apparatus 10. The face matching device 10 shown in the figure not only removes the influence of “illumination in a specific direction” as in the conventional face matching device, but also dynamically and adaptively changes the illumination even if the lighting direction and illuminance change. The face matching process is performed while removing the influence of the above.

  Specifically, the face collation apparatus 10 performs a learning process using learning data prior to a registration process of registration data and a collation process using input data. This learning data includes learning data with illumination variation and learning data without illumination variation. The learning data with illumination variation is face image data of a plurality of persons captured in illumination environments having different illumination directions and illuminances. The learning data without illumination fluctuation is a collection of a plurality of people's face image data of the same person that generated the learning data with illumination fluctuation and captured under a uniform illumination environment.

  The face collation apparatus 10 generates illumination influence reference data 13a, which will be described later, using learning data with illumination variation and learning data without illumination variation. The illumination influence reference data 13a is data serving as a reference when calculating the influence degree of illumination in the face image.

  The illumination influence reference data 13a indicates which of the facial parts is easily affected by the illumination and which is less susceptible to the illumination. For example, in the case of a part such as a chin line, even if there is an influence of illumination, the edge intensity only changes, and the edge itself remains, so that individual characteristics can be shown. For this reason, parts, such as a chin line, are parts which are comparatively hard to receive the influence of illumination. On the other hand, parts such as the nose and mouth that have a small difference in pixel value from the surrounding parts may become unclear depending on the illumination, making it difficult to show individual features. For this reason, the nose, mouth, and the like are parts that are relatively susceptible to illumination.

  In this way, if the illumination influence reference data 13a is generated in the learning stage, the face collation apparatus 10 accepts input of registration data including the face image of the person registered in advance or the feature amount of the face image, A feature amount in a plurality of local regions (segmented regions) of registration data is calculated. In addition, using the illumination influence reference data 13a generated from the learning data, a feature variation rate indicating how much the feature amount in each local region varies due to the influence of illumination is calculated.

  Next, when the face collation apparatus 10 receives input data composed of face image data of the person to be collated, the face collation apparatus 10 calculates feature quantities in a plurality of local regions of the input data. In addition, using the illumination influence reference data 13a generated from the learning data, a feature variation rate indicating how much the feature amount of each local region varies due to the illumination effect is calculated.

  Then, the face matching device 10 calculates the region weight in each local region using the feature variation rate of the registered data and the feature variation rate of the input data. The area weight is set to a smaller value as the feature variation rate is larger (the influence of illumination is larger).

  If the area weight is calculated in this way, the face matching device 10 calculates the similarity of the feature quantity in the corresponding local area between the registered data and the input data, and applies the area weight in the local area to the calculated similarity. Thus, the weighted similarity is calculated, and the registered data and the input data are collated using the weighted similarity. For example, if the sum of the weighted similarities is equal to or greater than a predetermined threshold, it is determined that they are the same person, and if it is less than the threshold, it is determined that they are different persons.

  As described above, the face collation apparatus 10 generates the illumination influence reference data 13a serving as a reference when calculating the illumination influence degree in the face image from the learning data, and the feature variation rate in each local region of the registration data and the input data. Is calculated from the illumination influence reference data 13a, and the matching process using the weighted similarity according to the feature variation rate is performed.

  The face collation apparatus 10 uses face image data of a plurality of people captured in various illumination environments having different illumination directions and illuminances as learning data, so that the face collation device 10 can be used not only in a specific direction but also in various illumination environments. Illumination influence reference data 13a indicating how much each local region corresponding to each part is susceptible to illumination is generated.

  Using this illumination influence reference data 13a, the feature variation rate in each local area of the registration data and the input data, that is, the magnitude of the influence by the illumination is evaluated, and the weight of the local area corresponding to the part having the large influence of the illumination is lowered. Thus, even if the lighting environment in the registration data and the input data is unknown, the lighting influence can be reduced dynamically and adaptively.

  Next, the configuration of the face matching apparatus 10 shown in FIG. 1 will be described using the block diagram of FIG. As illustrated in FIG. 2, the face collation apparatus 10 includes an image input receiving unit 11, a display operation unit 12, a storage unit 13, and a control unit 14.

  The image input receiving unit 11 is an interface that receives a face image as learning data, a registrant's face image as registration data, and a face collator's face image as input data from a camera, a scanner, or the like. The display operation unit 12 is an interface such as a touch panel display, a keyboard, and a mouse.

  The storage unit 13 is a storage device including a flash memory, a hard disk device, and the like, and stores illumination influence reference data 13a and registered feature data 13b.

  The illumination influence reference data 13a is generated and stored in the storage unit 13 when learning data is input. The registered feature data 13b is data including the feature amount 13c and the feature variation rate 13d of each local region of the registered data, and is generated and stored in the storage unit 13 when the registered data is input.

  The control unit 14 is a control unit that controls the face collating apparatus 10 as a whole. The control unit 14 includes an illumination influence reference data generation unit 14a, a feature amount calculation unit 14b, a feature variation rate calculation unit 14c, a similarity calculation unit 14d, a region weight calculation unit 14e, and a matching processing unit 14f.

  The illumination influence reference data generation unit 14 a is a processing unit that generates the illumination influence reference data 13 a from the learning data and stores it in the storage unit 13 when the image input reception unit 11 receives the learning data.

  The feature amount calculation unit 14b is a processing unit that calculates feature amounts for a plurality of local regions of received data when the image input reception unit 11 receives registration data or input data.

  When the image input reception unit 11 receives registration data or input data, the feature variation rate calculation unit 14c uses the illumination influence reference data 13a, and how much the feature amount of each local region varies due to the influence of illumination. It is a processing part which calculates the characteristic fluctuation rate which shows.

  The similarity calculation unit 14d is a processing unit that calculates the feature value similarity for the corresponding local regions of the registration data and the input data. The region weight calculation unit 14e calculates a region weight for each local region using the feature variation rate of each local region of the registered data and the feature variation rate of the local region of the input data. The area weight is set to a smaller value as the feature variation rate is larger, that is, as the influence of illumination is larger.

  The matching processing unit 14f weights each local region by multiplying the similarity of each local region calculated by the feature variation rate calculating unit 14c with the region weight of each local region calculated by the region weight calculating unit 14e. It is a processing unit that calculates a similarity and performs a matching process between registered data and input data using the weighted similarity for each local region.

  Next, generation of the illumination influence reference data 13a will be further described. FIG. 3 is an explanatory diagram for explaining generation of the illumination influence reference data 13a. The illumination influence reference data 13a includes an illumination variation vector 33, a feature variation coefficient matrix 32, and a feature amount distribution 31. The illumination influence reference data generation unit 14a generates an illumination variation vector 33, a feature variation coefficient matrix 32, and a feature amount distribution 31 from learning data with illumination variation and learning data without feature variation.

  The illumination influence reference data generation unit 14a first reduces the resolution of the learning data with illumination variation. For example, in the learning data of 80 pixels in the vertical direction and 80 pixels in the horizontal direction, the value of the number of pixels “Q” is “80 × 80 = 6400”. When the resolution of the learning data is reduced to reduced data of 10 pixels in the vertical direction and 10 pixels in the horizontal direction, the value of the number of pixels “q” in the reduced data is “10 × 10 = 100”. As a specific technique for reducing the resolution, any method such as smoothing, averaging of pixel values, and thinning can be used.

  In this way, by reducing the resolution of the learning data for the number of pixels “Q” to the number of pixels “q”, the components corresponding to the individual characteristics are reduced, and the components corresponding to the influence of illumination remain. Therefore, when reduced data is generated for r pieces of learning data having different illumination environments, r pieces of reduced data corresponding to the respective illumination environments are obtained.

  The illumination influence reference data generation unit 14a generates a pixel value matrix of r rows and q columns using r pieces of reduced data with the number of pixels q. In this pixel value matrix, q elements in one row are pixel values of q pixels in one reduced data.

  The illumination influence reference data generation unit 14a obtains a covariance matrix of a pixel value matrix of r rows and q columns. Then, eigenvalues of this covariance matrix are calculated, and k eigenvalues are selected in descending order. The illumination influence reference data generation unit 14 a sets the k eigenvectors corresponding to the selected eigenvalue as the illumination variation vector 33. An arbitrary number can be used as the value of k. The illumination variation vector 33 is used as illumination environment reference data that serves as a reference when obtaining illumination parameters corresponding to the illumination environment from a plurality of face image data having different illumination environments. Here, the illumination parameter indicates the direction of illumination obtained from the captured image.

  With this illumination variation vector 33, an illumination parameter corresponding to the illumination environment of the reduced data can be obtained. Specifically, the illumination influence reference data generation unit 14a uses the q pixel values of the reduced data as a 1-row q-column matrix and obtains a product with the illumination variation vector 33, whereby the illumination parameter X of the reduced data is obtained. Is calculated. By calculating the illumination parameter X for r pieces of reduced data, r illumination parameters X are obtained.

  In addition, the illumination influence reference data generation unit 14a calculates the feature amounts of n local regions P1 to Pn in the learning data with illumination fluctuation before performing the resolution reduction at the time of learning. The feature amount is assumed to be m-dimensional. For example, when 5 × 5 sample points are set around one local region and the x-direction differential value and y-direction differential value of each sample point are taken, m = 5 × 5 × 2 = 50 dimensions It becomes.

  At the time of learning, the illumination influence reference data generation unit 14a calculates feature amounts of the local regions P1 to Pn for each of the r pieces of learning data. Then, an average feature amount for each individual is calculated for each of the local regions P1 to Pn. Then, for each of the r pieces of learning data, a deviation between the feature amount of the local regions P1 to Pn and the average feature amount of the corresponding individual region of the corresponding individual is calculated as a feature variation amount Y1 to Yn. As a result, r fluctuation amounts Y1 to Yn of the local regions P1 to Pn are obtained. Since the feature amount is m-dimensional, the feature variation amounts Y1 to Yn are also m-dimensional.

The illumination influence reference data generation unit 14a uses the feature variation amounts Y1 to Yn obtained from the same learning data and the illumination parameter X,
Y1 = A1X, ... Yn = AnX
To calculate the coefficient matrix A1,... An. The coefficient matrices A1,..., An are m rows and k columns, and the coefficient matrices A1 to An are obtained corresponding to the local regions P1 to Pn. The coefficient matrixes A1 to An are the feature variation coefficient matrix 32, and are used as feature variation reference data serving as a reference when obtaining the feature variation amount of the local region under the illumination environment corresponding to the illumination parameter.

  Furthermore, the illumination influence reference data generation unit 14a calculates a feature amount distribution from learning data without illumination fluctuation. Specifically, the illumination influence reference data generation unit 14a obtains the feature amount of each local area from the learning data of a plurality of persons without illumination variation captured in a uniform illumination environment, and calculates the feature amount for each local area. Find the distribution. Then, the distribution width of each dimension of the feature amount is summed to obtain a feature amount distribution 31 of the local region.

  The face collation apparatus 10 uses the feature amount distribution 31 for calculating the feature variation rate of the registration data and the input data. FIG. 4 is an explanatory diagram for explaining the feature amount distribution 31 and the feature variation amount. FIG. 4A shows a first-dimensional distribution and a second-dimensional distribution among the feature amount distributions. In FIG. 4A, the feature amount in the case where there is no illumination variation, the width of the first-dimensional distribution (variation amount) is L1, and the width of the second-dimensional distribution is L2. Although FIG. 4A illustrates two dimensions, there is a distribution width corresponding to the dimension number m of the feature amount. As shown in FIG. 4B, the feature amount in the case where there is an illumination variation has a first-dimensional distribution width (variation amount) of 11 and a second-dimensional distribution width of 12. However, l1 and l2 differ depending on illumination conditions. The calculation of the feature variation amount is performed by a feature variation amount calculation unit 53 inside the feature variation rate calculation unit 14c described later.

  Next, details of the illumination influence reference data generation unit 14a illustrated in FIG. 2 will be described. FIG. 5 is an explanatory diagram for describing the details of the illumination influence reference data generation unit 14a illustrated in FIG. As shown in FIG. 5, the illumination influence reference data generation unit 14a includes therein a feature amount calculation unit 41, a feature amount distribution calculation unit 42, a feature variation calculation unit 43, a resolution reduction processing unit 44, and an illumination variation vector calculation unit. 45, an illumination parameter calculation unit 46, and a feature variation coefficient matrix calculation unit 47.

  The feature amount calculation unit 41 is a processing unit that calculates feature amounts for a plurality of local regions of learning data. The feature amount calculation unit 41 outputs the calculated feature amount to the feature amount distribution calculation unit 42 when the feature amount is calculated for each local region of the learning data without illumination variation. Further, when the feature amount is calculated for each local region of the learning data with illumination variation, the feature amount calculation unit 41 outputs the calculated feature amount to the feature variation amount calculation unit 43.

  The feature amount distribution calculation unit 42 obtains the distribution of feature amounts calculated by the feature amount calculation unit 41 from the learning data of a plurality of persons without illumination variation for each local region. Then, the distribution width of each dimension of the feature amount is summed to obtain a feature amount distribution 31 of the local region.

  The feature fluctuation amount calculation unit 43 calculates the average feature amount of each local region for each individual using the feature amount calculated from the plurality of learning data with illumination fluctuation by the feature amount calculation unit 41. Then, the feature variation amount calculation unit 43 calculates, as a feature variation amount Y, the deviation of the feature amount of each local region from the individual average feature amount for a plurality of image data with illumination variation, and calculates a feature variation coefficient matrix To the unit 47.

  The resolution reduction processing unit 44 is a processing unit that reduces the resolution of learning data with illumination fluctuation. The resolution reduction processing unit 44 outputs reduced data obtained by reducing the resolution of the learning data with illumination variation to the illumination variation vector calculation unit 45 and the illumination parameter calculation unit 46.

  The illumination variation vector calculation unit 45 uses r pieces of reduced data with the number of pixels q, generates a pixel value matrix of r rows and q columns, and obtains a covariance matrix of the pixel value matrix. Then, k eigenvalues of this covariance matrix are selected in descending order, and k eigenvectors corresponding to the selected eigenvalues are set as illumination variation vectors 33. The illumination variation vector calculation unit 45 outputs the illumination variation vector 33 to the illumination parameter calculation unit 46.

  The illumination parameter calculation unit 46 calculates the illumination parameter X of the reduced data by obtaining a product of the q pixel values of the reduced data as a 1-row q-column matrix and the illumination variation vector 33. By calculating the illumination parameter X for r pieces of reduced data, r illumination parameters X are obtained. The illumination parameter calculation unit 46 outputs the r illumination parameters X to the feature variation coefficient matrix calculation unit 47.

The feature variation coefficient matrix calculation unit 47 uses the feature variation amount Y calculated by the feature variation amount calculation unit 43 and the illumination parameter X calculated by the illumination parameter calculation unit 46.
Y1 = A1X, ... Yn = AnX
To calculate the coefficient matrix A1,... An. The coefficient matrices A1,..., An are m rows and k columns, and the coefficient matrices A1 to An are obtained corresponding to the local regions P1 to Pn. The feature variation coefficient matrix calculation unit 47 sets the coefficient matrices A1 to An as the feature variation coefficient matrix 32.

  The illumination influence reference data generation unit 14a includes the feature amount distribution 31 calculated by the feature amount distribution calculation unit 42, the feature variation coefficient matrix 32 calculated by the feature variation coefficient matrix calculation unit 47, and the illumination calculated by the illumination variation vector calculation unit 45. The variation vector 33 is stored in the storage unit 13 as the illumination influence reference data 13a.

  Next, generation of the feature amount 13c and the feature variation rate 13d included in the registered feature data 13b will be further described. FIG. 6 is an explanatory diagram for explaining generation of the feature amount 13c and the feature variation rate 13d.

  The feature amount calculation unit 14b calculates the feature amount 13c of the local regions P1 to Pn from the registered data. The feature variation rate calculation unit 14c calculates the feature variation rate 13d of the local regions P1 to Pn from the registered data.

  The feature variation rate calculation unit 14c first reduces the resolution of the registered data. For example, in the registration data of 80 pixels vertically and 80 pixels horizontally, the value of the number of pixels “Q” is “80 × 80 = 6400”. When the resolution of the registered data is reduced to reduced data of 10 pixels in the vertical direction and 10 pixels in the horizontal direction, the value of the number of pixels “q” in the reduced data is “10 × 10 = 100”. As a specific technique for reducing the resolution, any method such as smoothing, averaging of pixel values, and thinning can be used.

  As described above, by reducing the resolution of the registration data of the number of pixels “Q” to the number of pixels “q”, the component corresponding to the personal feature is reduced and the component corresponding to the influence of illumination remains.

  The feature variation rate calculation unit 14c calculates the illumination parameter X of the reduced data by determining the product of the q pixel values of the reduced data as a 1-row q-column matrix and the illumination variation vector 33.

The feature variation rate calculation unit 14c performs linear conversion based on the illumination parameter X and the feature variation coefficient matrix 32, that is,
Y = AX
Thus, the feature fluctuation amount Y is calculated. The feature variation coefficient matrix 32 has coefficient matrices A1 to An corresponding to the local regions P1 to Pn. The coefficient matrices A1 to An are each a matrix of m rows and k columns. By obtaining the product of the illumination parameter X and the coefficient matrices A1 to An, the feature variation amounts Y1 to Yn corresponding to the local regions P1 to Pn can be obtained.

  The feature variation rate calculation unit 14 c calculates the feature variation rate rY by dividing the sum of the values of each dimension of the feature variation amount Y by the feature amount distribution 31. The feature variation rate rY is obtained for each local region. The feature variation rate calculation unit 14c sets the feature variation rates rY1 to rYn corresponding to the local regions P1 to Pn as the feature variation rate 13d.

  Next, details of registration and verification will be described. FIG. 7 is an explanatory diagram for explaining details of registration and collation. As shown in FIG. 7, when the feature amount calculation unit 14b calculates the feature amount of the local region from the registration data, the feature amount calculation unit 14b sets the calculated feature amount as the feature amount 13c of the registration feature data 13b. In addition, when the feature amount calculation unit 14b calculates the feature amount of the local region from the input data, the feature amount calculation unit 14b outputs the calculated feature amount to the similarity calculation unit 14d.

  When calculating the feature variation rate of the local region from the registered data, the feature variation rate calculating unit 14c sets the calculated feature variation rate of the local region as the feature variation rate 13d of the registered feature data 13b. In addition, when the feature variation rate of the local region is calculated from the input data, the feature variation rate calculation unit 14c outputs the calculated feature variation rate of the local region to the region weight calculation unit 14e.

Here, the feature variation rate is obtained by dividing the sum of the values of each dimension of the feature variation amount by the feature amount distribution when there is an illumination variation. In FIG. 4A, the distribution of the feature quantity is L1 in the first dimension and L2 in the second dimension. In FIG. 4B, the first dimension of the feature variation is l1, and the second dimension is l2. In this case, the feature variation rate rY is
rY = (l1 + l2 +... + ln) / (L1 + L2 +... + L50)
Is required.

  The similarity calculation unit 14d calculates, for each local region, the similarity with the feature value 13c of the registered feature data 13b when the feature value of the input data is output by the feature value calculation unit 14b. The similarity calculation unit 14d outputs the calculated similarity for each local region to the matching processing unit 14f.

  When the feature variation rate of the input data is output from the feature variation rate computation unit 14c, the region weight calculation unit 14e calculates the region weight using the feature variation rate 13d of the registered feature data 13b and the feature variation rate of the input data. Calculate and output to the verification processing unit 14f.

  The matching processing unit 14f calculates the weighted similarity by multiplying the similarity for each local region calculated by the similarity calculation unit 14d by the weight for each local region calculated by the region weight calculation unit 14e. Registration data and input data are collated using the weighted similarity. Specifically, the matching processing unit 14f calculates the sum of the weighted similarities, and determines that the registered data and the input data are the same person when the sum of the weighted similarities is equal to or greater than a threshold value.

  The feature variation rate calculation unit 14c includes a resolution reduction processing unit 51, an illumination parameter calculation unit 52, a feature variation amount calculation unit 53, and a variation rate calculation processing unit 54 therein. The resolution reduction processing unit 51 is a processing unit that reduces the resolution of registration data and input data. The resolution reduction processing unit 51 outputs reduced data obtained by reducing the resolution to the illumination parameter calculation unit 52.

  The illumination parameter calculation unit 52 uses the q pixel values of the reduced data as a matrix of 1 row and q column, and obtains the illumination parameter X of the reduced data by obtaining a product with the illumination variation vector 33 read from the storage unit 13. calculate. The illumination parameter calculation unit 52 outputs the calculated illumination parameter X to the feature fluctuation amount calculation unit 53.

  The feature variation calculation unit 53 is a processing unit that calculates the feature variation Y by linear conversion using the illumination parameter X and the feature variation coefficient matrix 32 read from the storage unit 13. The feature variation amount calculation unit 53 outputs the calculated feature variation amount Y to the variation rate calculation processing unit 54. The variation rate calculation processing unit 54 calculates the feature variation rate rY by dividing the sum of the values of each dimension of the feature variation amount Y by the feature amount distribution 31.

  Next, calculation of region weights will be described. As the area weight, any function such as a logistic sigmoid function, a quadratic function, or a cubic function can be used as long as it is a function that monotonously decreases depending on the value of the feature variation rate.

  FIG. 8 is an explanatory diagram for explaining a case where the logistic sigmoid function is used for calculating the region weight. The logistic sigmoid function is a point-symmetrical function with respect to (x, y) = (0, 0.5). If the feature variation rate is the x axis and the region weight is the y axis, the feature variation rate increases. The area weight becomes smaller.

  Next, a processing procedure related to learning and registration of the face matching device 10 will be described. FIG. 9 is a flowchart illustrating a processing procedure related to learning and registration of the face matching device 10.

  First, the image input receiving unit 11 receives learning data (step S101). The illumination influence reference data generation unit 14a performs illumination influence reference data generation processing using the learning data (Step S102), and stores the generated illumination influence reference data 13a in the storage unit 13 (Step S103).

  When the image input receiving unit 11 receives input of registration data (step S104), the feature amount calculation unit 14b calculates feature amounts for a plurality of local regions of the registration data (step S105).

  Further, the feature variation rate calculation unit 14c performs a process of calculating the feature variation rate using the registration data received by the image input reception unit 11 and the illumination influence reference data 13a (step S106). Thereafter, the face matching device 10 stores the feature amount calculated in step S105 and the feature variation rate calculated in step S106 in the storage unit 13 as registered feature data 13b (step S107), and ends the process.

  Next, the illumination influence reference data generation process shown as step S102 in FIG. 9 will be described. FIG. 10 is a flowchart showing the processing procedure of the illumination influence reference data generation processing.

  The feature amount calculation unit 41 of the illumination influence reference data generation unit 14a first calculates feature amounts for a plurality of local regions of learning data without illumination variation (step S201).

  The feature quantity distribution calculation unit 42 obtains the distribution of feature quantities calculated by the feature quantity calculation unit 41 from the learning data of a plurality of persons without illumination fluctuation for each local region, sums the distribution width of each dimension of the feature quantities, The quantity distribution 31 is set (step S202).

  The feature amount calculation unit 41 calculates feature amounts for a plurality of local regions of the learning data with illumination variation (step S203). The feature variation calculation unit 43 uses the feature amount calculated by the feature calculation unit 41 from a plurality of learning data with illumination variation, calculates an average feature amount for each local region, and calculates the feature amount of each local region. Is calculated as a feature variation amount Y (step S204).

  The resolution reduction processing unit 44 reduces the resolution of the learning data with illumination variation (step S205). The illumination variation vector calculation unit 45 uses r pieces of reduced data with the number of pixels q, generates a pixel value matrix of r rows and q columns, and obtains a covariance matrix of the pixel value matrix. Then, k eigenvalues of the covariance matrix are selected in descending order, and k eigenvectors corresponding to the selected eigenvalues are calculated as the illumination variation vector 33 (step S206).

  The illumination parameter calculation unit 46 calculates q pixel values of the reduced data as a 1-row q-column matrix and calculates a product with the illumination variation vector 33, thereby calculating the illumination parameter X of the reduced data (step S207). .

The feature variation coefficient matrix calculation unit 47 uses the feature variation amount Y calculated in step S204 and the illumination parameter X calculated in step S207.
Y = AX
Thus, the feature variation coefficient matrix 32 is obtained.

  The illumination influence reference data generation unit 14a uses the feature amount distribution 31 calculated in step S202, the feature variation coefficient matrix 32 calculated in step S208, and the illumination variation vector 33 calculated in step S206 as illumination influence reference data 13a. (Step S209), the process ends.

  Next, the feature variation rate calculation process shown as step S106 in FIG. 9 will be described. FIG. 11 is a flowchart illustrating the processing procedure of the feature variation rate calculation processing.

  The resolution reduction processing unit 51 of the feature variation rate calculation unit 14c performs a process of reducing the resolution of registration data or input data and generating reduced data (step S301). The illumination parameter calculation unit 52 uses the q pixel values of the reduced data as a matrix of 1 row and q column, and obtains the illumination parameter X of the reduced data by obtaining a product with the illumination variation vector 33 read from the storage unit 13. Calculate (step S302).

  The feature variation amount calculation unit 53 calculates the feature variation amount Y by linear conversion using the illumination parameter X and the feature variation coefficient matrix 32 read from the storage unit 13 (step S303). The variation rate calculation processing unit 54 calculates the feature variation rate rY by dividing the sum of the values of each dimension of the feature variation amount Y by the feature amount distribution 31 (step S304), and ends the processing.

  Next, a processing procedure related to collation of the face collation apparatus 10 will be described. FIG. 12 is a flowchart illustrating a processing procedure related to the collation of the face collation apparatus 10.

  First, the image input receiving unit 11 receives input data (step S401). The feature amount calculation unit 14b calculates feature amounts for a plurality of local regions from the input data (step S402).

  Further, the feature variation rate calculation unit 14c performs a feature variation rate calculation process for calculating a feature variation rate using the input data received by the image input reception unit 11 and the illumination influence reference data 13a (step S403). As the feature variation rate calculation process, the process shown in FIG. 11 may be used.

  Thereafter, the face collation apparatus 10 selects the registered feature data 13b to be collated with the input data (step S404). The region weight calculation unit 14e compares the feature variation rate of the selected registered feature data 13b with the feature variation rate of the input data for each local region, and selects the one with the larger feature variation rate (step S405). The area weight calculation unit 14e calculates the area weight of each local area based on the selected feature variation rate (step S406).

  The similarity calculation unit 14d calculates the similarity between the feature quantity of the selected registered feature data 13b and the feature quantity of the input data for each local region (step S407). The matching processing unit 14f calculates the weighted similarity by multiplying the similarity calculated in step S407 by the region weight calculated in step S406, and totals the weighted similarities (step S408).

  If the total weighted similarity (total similarity) is equal to or greater than the threshold (step S409; Yes), the matching processing unit 14f determines that the selected registered feature data 13b and the input data belong to the same person. Then, information (name and the like) related to the selected registered feature data 13b is output as a collation result (step S412), and the process ends.

  If the total similarity is less than the threshold (step S409; No), the collation processing unit 14f determines whether all the registered feature data 13b have been selected (step S410).

  If unselected registered feature data 13b remains (step S410; No), the matching processing unit 14f selects the registered feature data 13b to be matched with the input data from the unselected registered feature data 13b (step S404). ). If all the registered feature data 13b have been selected (step S410; Yes), the collation processing unit 14f outputs “no corresponding person” as the collation result, and ends the process.

  Next, imaging of learning data will be described. When imaging learning data with illumination variation, a human face is imaged in an illumination environment with different illumination directions and illuminance. FIG. 13 is an explanatory diagram for describing imaging of learning data.

  As shown in FIG. 13, a plurality of LED (Light Emitting Diode) lights 71 are installed around a person who is a subject, and an imaging device 61 is provided in front of the person who is the subject. The imaging device 61 is controlled in imaging timing by the imaging control device 60 and transmits an imaging result to the imaging control device 60.

  The imaging control device 60 is connected to the illumination control device 70 and can transmit an illumination control instruction to the illumination control device 70. The illumination control device 70 realizes an illumination environment according to the illumination control instruction by individually controlling the lighting state of the LED illumination 71.

  For this reason, the imaging control device 60 can virtually realize various illumination environments having different illumination directions and illuminances depending on the lighting state of the LED illumination 71, and can capture learning data under each illumination environment.

  Next, the case where each function part of the face collation apparatus 10 shown in FIG. 2 is implement | achieved by a program is demonstrated. FIG. 14 is an explanatory diagram in the case where each functional unit of the face collation apparatus shown in FIG. 2 is realized by a program. As shown in the figure, the face collation apparatus 10 shown in FIG. 2 has a configuration in which a CPU 111 (including a main memory), a flash memory 112, a display unit 113, and an operation reception unit 114 are connected by a bus 119.

  Here, the face matching program 130 is stored in the flash memory 112, and the CPU 111 reads out the face matching program 130 from the flash memory 112 and expands it in a main memory (not shown), and executes the process of the face matching program 130. To do.

  Specifically, the face collation program 130 includes an illumination influence reference data generation process 120a, a feature amount calculation process 120b, a feature variation rate calculation process 120c, a similarity calculation process 120d, an area weight calculation process 120e, and a collation process 120f. Generated.

  The illumination influence reference data generation process 120a corresponds to the illumination influence reference data generation unit 14a shown in FIG. 2, and the feature amount calculation process 120b corresponds to the feature amount calculation unit 14b shown in FIG. 120c corresponds to the feature variation rate calculation unit 14c shown in FIG. 2, the similarity calculation process 120d corresponds to the similarity calculation unit 14d shown in FIG. 2, and the region weight calculation process 120e uses the region weight shown in FIG. Corresponding to the calculation unit 14e, the collation process 120f corresponds to the collation processing unit 14f shown in FIG.

  As described above, the face collation apparatus 10 according to the present embodiment generates the illumination influence reference data 13a serving as a reference when calculating the illumination influence degree in the face image from the learning data, and stores the registration data and the input data. The feature variation rate in each local area is calculated from the illumination influence reference data 13a, and the matching process is performed based on the weighted similarity according to the feature variation rate. For this reason, even if the illumination environment in registration data and input data is unknown, the illumination influence can be reduced dynamically and adaptively.

  In the above embodiment, when the total similarity between any registered feature data and input data is equal to or greater than the threshold value, the process is terminated and the total similarity with any registered feature data is also set. The notification is also performed when the threshold value is less than the threshold value, but how to output the collation result can be appropriately set.

  For example, the total similarity may be calculated for all the registered feature data, and information related to the registered feature data having the maximum total similarity may be output as a matching result. Further, all registered feature data whose total similarity exceeds a threshold value may be output as a matching result. Further, when the total similarity with any registered feature data is less than the threshold value, notification is not necessary.

  Further, in the above-described embodiment, the case where the matching is performed based on the total similarity obtained by summing all the weighted similarities weighted by the feature variation rate has been described. However, the similarity of a local region having a small feature variation rate is described. The matching process may be performed using selectively.

  In the above-described embodiment, the case where the feature variation rate is obtained for both the input data and the registered data and one of them is used to calculate the weight for the similarity is described. However, the feature variation rate of the input data and the input data are described. The weight for the similarity may be calculated using both of the feature variation rates. Alternatively, the feature variation rate may be calculated for only one of the input data and the registered data, and the weight for the similarity may be calculated. For example, if the registered data can be imaged in a uniform illumination environment, the feature variation rate may be calculated only for the input data, and the weight for the similarity may be calculated.

  As described above, the face matching device, the face matching method, and the face matching program according to the present invention are suitable for dynamically and adaptively reducing the influence of the light source.

DESCRIPTION OF SYMBOLS 10 Face collation apparatus 11 Image input reception part 12 Display operation part 13 Memory | storage part 13a Illumination influence reference data 13b Registration feature data 13c Feature amount 13d Feature fluctuation rate 14 Control part 14a Illumination influence reference data generation part 14b Feature quantity calculation part 14c Feature fluctuation Rate calculation unit 14d Similarity calculation unit 14e Region weight calculation unit 14f Collation processing unit 31 Feature amount distribution 32 Feature variation coefficient matrix 33 Illumination variation vector 41 Feature amount calculation unit 42 Feature amount distribution calculation unit 43, 53 Feature variation amount calculation unit 44 , 51 Low resolution processing unit 45 Illumination variation vector calculation unit 46, 52 Illumination parameter calculation unit 47 Feature variation coefficient matrix calculation unit 54 Variation rate calculation processing unit 60 Imaging control device 61 Imaging device 70 Lighting control device 71 LED illumination

Claims (10)

A face collation device for collating input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a face image registered beforehand or its feature quantity,
Illumination influence reference data calculating means for calculating illumination influence reference data serving as a reference when calculating the influence degree of the light source in the face image using learning data including a plurality of face images having different illumination environments ;
Based on the illumination influence reference data calculated by the illumination influence reference data calculation means, an influence degree calculation means for calculating the influence degree of the light source in each local region of the input data and / or the registration data;
Similarity calculating means for calculating a similarity between a feature amount in each local region of the input data and a feature amount in a corresponding local region of the registration data;
A collation process for collating the input data and the registered data based on the similarity between the feature amounts in each local region calculated by the similarity calculation unit and the influence calculated by the influence calculation unit and means,
The lighting influence reference data calculation means includes lighting environment reference data serving as a reference when obtaining lighting parameters corresponding to the lighting environment from the learning data including a plurality of face images having different lighting environments, and the lighting parameters. Feature variation reference data serving as a reference for obtaining a feature variation amount generated in the feature amount in the local region under the corresponding illumination environment is calculated as the illumination influence reference data.
Face matching device comprising a call. Before Symbol influence calculation means, said calculating illumination parameters of the input data and / or the registration data using the illumination environment reference data, the input data and using the calculated and illumination parameters and the characteristic variation reference data / Or calculate a feature variation rate in each local region of the registration data,
Further comprising region weight calculating means for calculating the weight of each local region according to the feature variation rate;
The collation processing unit performs a collation process on the input data and the registration data by weighting the similarity of each local region using the region weight calculated by the region weight calculation unit. Item 2. The face matching device according to Item 1. The illumination influence reference data calculation means performs a resolution reduction process for reducing the resolution for the learning data including a plurality of face images having different illumination environments, and the learning data subjected to the resolution reduction process performs the resolution reduction process. Calculate lighting environment standard data,
The influence calculation means performs a resolution reduction process for reducing the resolution of the input data and / or the registration data, and the input data and / or the registration data and the illumination subjected to the resolution reduction process. The face verification device according to claim 2, wherein the illumination parameter is calculated using environmental reference data. The illumination influence reference data calculation means is a pixel value matrix in which the number of pixels of the learning data subjected to the resolution reduction processing is the number of rows or columns, and the number of the plurality of face image data is the number of columns or rows. Generating the eigenvector of the pixel value matrix as the lighting environment reference data,
The influence degree calculating means calculates a product of each pixel value of the input data and / or the registration data subjected to the resolution reduction processing and the illumination environment reference data as the illumination parameter. Item 4. The face matching device according to Item 3. The illumination influence reference data calculation means calculates a product of each pixel value of the face image of the learning data subjected to the resolution reduction process and the illumination environment reference data as the illumination parameter, and the resolution reduction process The feature variation amount of each local region is calculated from the previous learning data, and for each local region, the linear transformation coefficient matrix when the local region feature variation amount is assumed to be a linear transformation of the illumination parameter Calculated as feature variation reference data,
The influence degree calculating means calculates the product of the illumination parameter calculated from the input data and / or the registration data subjected to the resolution reduction process and the feature variation reference data as the input data and / or the registration data. The face matching device according to claim 4, wherein the face matching device is calculated as a feature variation amount.   The area weight calculating unit uses a larger characteristic variation rate for each local region when the influence degree calculating unit calculates a characteristic variation rate in each local region for both the input data and the registered data. 6. The face collating apparatus according to claim 2, wherein the area weight is calculated. A face collation method for collating input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a face image registered beforehand or its feature quantity,
Illumination influence reference data calculation step for calculating illumination influence reference data serving as a reference when calculating the influence degree of the light source in the face image using learning data including a plurality of face images with different illumination environments ;
Based on the illumination influence reference data calculated by the illumination influence reference data calculation step, an influence calculation step for calculating the influence degree of the light source in each local region of the input data and / or the registration data;
A similarity calculation step of calculating a similarity between a feature amount in each local region of the input data and a feature amount in a corresponding local region of the registration data;
A collation process for collating the input data with the registered data based on the similarity between the feature amounts in each local region calculated in the similarity calculation step and the influence calculated in the influence calculation step and a step seen including,
In the lighting influence reference data calculating step, the lighting environment reference data serving as a reference for obtaining the lighting parameter corresponding to the lighting environment from the learning data including a plurality of face images having different lighting environments, and the lighting parameter Feature variation reference data serving as a reference for obtaining a feature variation amount generated in the feature amount in the local region under the corresponding illumination environment is calculated as the illumination influence reference data.
Face matching wherein a call. A face collation program for collating input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a face image registered beforehand or its feature quantity,
Illumination influence reference data calculation procedure for calculating illumination influence reference data to be a reference when calculating the influence degree of the light source in the face image using learning data including a plurality of face images with different illumination environments ;
Based on the illumination influence reference data calculated by the illumination influence reference data calculation procedure, an influence calculation procedure for calculating an influence degree of a light source in each local region of the input data and / or the registration data;
A similarity calculation procedure for calculating a similarity between a feature amount in each local region of the input data and a feature amount in a corresponding local region of the registration data;
A collation process for collating the input data with the registered data based on the similarity of the feature amount in each local area calculated by the similarity calculation procedure and the influence calculated by the influence calculation procedure Let the computer perform the steps and
In the lighting influence reference data calculation procedure, the lighting environment reference data serving as a reference for obtaining the lighting parameter corresponding to the lighting environment from the learning data including a plurality of face images having different lighting environments, and the lighting parameter Feature variation reference data serving as a reference for obtaining a feature variation amount generated in the feature amount in the local region under the corresponding illumination environment is calculated as the illumination influence reference data.
Face recognition program, wherein a call.   A face collation device for collating input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a face image registered beforehand or its feature quantity,
  The input data and / or the registration based on the illumination influence reference data, which is calculated in advance using learning data including a plurality of face images with different illumination environments and serves as a reference when calculating the influence degree of the light source in the face image An influence calculation means for calculating the influence degree of the light source in each local region of the data;
  Similarity calculating means for calculating a similarity between a feature amount in each local region of the input data and a feature amount in a corresponding local region of the registration data;
  A collation process for collating the input data and the registered data based on the similarity between the feature amounts in each local region calculated by the similarity calculation unit and the influence calculated by the influence calculation unit Means and
  With
  The illumination influence reference data includes illumination environment reference data serving as a reference for obtaining an illumination parameter corresponding to the illumination environment, and a feature variation amount generated in a feature amount in the local region under the illumination environment corresponding to the illumination parameter. Includes feature variation reference data that serves as a reference
  A face matching apparatus characterized by the above.   A face collation program for collating input data consisting of a face image of a person to be collated or its feature quantity with registered data consisting of a face image registered beforehand or its feature quantity,
  The input data and / or the registration based on the illumination influence reference data, which is calculated in advance using learning data including a plurality of face images with different illumination environments and serves as a reference when calculating the influence degree of the light source in the face image An influence calculation procedure for calculating the influence of the light source in each local region of the data;
  A similarity calculation procedure for calculating a similarity between a feature amount in each local region of the input data and a feature amount in a corresponding local region of the registration data;
  A collation process for collating the input data with the registered data based on the similarity of the feature amount in each local area calculated by the similarity calculation procedure and the influence calculated by the influence calculation procedure Procedure and
  To the computer,
  The illumination influence reference data includes illumination environment reference data serving as a reference for obtaining an illumination parameter corresponding to the illumination environment, and a feature variation amount generated in a feature amount in the local region under the illumination environment corresponding to the illumination parameter. Includes feature variation reference data that serves as a reference
  A face matching program characterized by that.

Images