JP4187494B2 - Image recognition apparatus, image recognition method, and program for causing computer to execute the method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
According to the present invention, an image is obtained by comparing local features at sample points between an input image obtained by capturing a predetermined part of a person's body such as a face and a palm print and a plurality of registered images of the person and others related to the part. The present invention relates to an image recognition apparatus that collates and confirms identity, an image recognition method, and a program that causes a computer to execute the method, and in particular, an image recognition apparatus that can efficiently identify a person's image and another person's image and increase the recognition rate. The present invention relates to an image recognition method and a program for causing a computer to execute the method.
[0002]
[Prior art]
Conventionally, in a security system related to admission management, access to a computer, and the like, a technique for identifying a person and others using the physical characteristics of an individual has been used. In the method of verifying the identity by comparing the input image with a registered image such as a face or palm print that has been registered in advance, the facial expression, the degree of opening of the palm, the shooting angle, etc. Due to the influence of the difference, it is not possible to exhibit high discrimination ability simply by superimposing the input image and a plurality of registered images and comparing the degree of matching.
[0003]
As an image recognition method for verifying the identity by comparing the input image with the registered image, the size of each image in the image set is normalized, and the feature vector subspace consisting of the gray value of each pixel of the normalized image is mainly used. An eigenface method is generally known that is generated by component analysis, projects feature vectors of an input image and a registered image onto a partial space, calculates similarity, and makes a collation determination.
[0004]
According to this conventional technique, since the input face image and the registered face image are collated based on each fixed pixel between the input image and the registered image, there is a difference in facial expression, shooting angle, etc. Thus, when the image is distorted or misaligned between the input face image and the registered face image, there is a problem that the ability to discriminate between the person and others falls.
[0005]
For this reason, sample points are designated at portions that are marks of the face such as the eyes, nose and mouth on the registered images of the face, and the positions of the sample points on the input image corresponding to the sample points are calculated for the similarity between the two images. And a technique for comparing local characteristics of both images between the obtained sample points is disclosed (for example, see Non-Patent Document 1).
[0006]
That is, in this prior art, even when distortion or deviation occurs between the registered image and the input image due to differences in facial expression or shooting angle, etc., it is possible to deal with sample points on the registered image by calculating similarity. Since the sample points on the input image to be detected can be detected, the above problems are alleviated, and the ability to discriminate between the person and others can be improved.
[0007]
[Non-Patent Document 1]
L. Wiskott, 3 others, “Face Recognition by Elastic Bunch Graph Matching”, Eye Triple E Transactions on Pattern・ Analysis and Machine Intelligence (IEEE Transactions On Pattern Analysis And Machine Intelligence), 1997, Vol. 19, No. 7, p. 775-779
[0008]
[Problems to be solved by the invention]
However, when this conventional technique (for example, refer to Non-Patent Document 1) is used, in order to perform collation between the input image and a plurality of registered images at the corresponding sample points, all the registered images are sampled in advance. It is necessary to specify points, and as the number of registered images increases, the burden of the sample point specifying work increases, and the storage capacity for storing the sample points of each registered image increases. .
[0009]
In addition, since the registered images such as faces and palm prints used for collation and the input image include features common to all people, it is difficult to clearly detect individual features, and it is difficult to improve identification ability. There is a problem.
[0010]
Furthermore, although it can deal with some distortions and shifts of the input image with respect to the registered image caused by differences in facial expressions, palm opening, shooting angle, etc., it can reduce fluctuations in the image of the same person and reduce the image of others There is a problem that it is difficult to further improve the discrimination capability because no processing that emphasizes the difference from the above is performed.
[0011]
The present invention has been made in view of the above problems (problems), and reduces the burden of setting sample points on each registered image, and also learns the characteristics of an image set to make those characteristics efficient. Providing an image recognition apparatus, an image recognition method, and a program for causing a computer to execute the method that can efficiently identify a person's image and another person's image by using well-represented knowledge and increase the recognition rate The purpose is to do.
[0012]
[Means for Solving the Problems]
In order to achieve the above object, an image recognition apparatus according to the first aspect of the present invention provides a sample between an input image obtained by capturing a predetermined part of a person's body and a plurality of registered images of the person and others relating to the part. In an image recognition apparatus that compares an image by comparing local features at points and performs identity verification, an average image creating unit that creates an average image from a plurality of images in an image set related to the part; Subspace generating means for detecting a sample point on each image in the image set corresponding to the set sample point and generating a subspace of a feature vector at the sample point; and a sample point on the average image Registered image sample point detecting means for detecting a corresponding sample point on the registered image using the partial space, and before corresponding to the sample point on the registered image Input image sample point detection means for detecting sample points on the input image using the partial space, feature vectors at the sample points on the registered image detected by the registered image sample point detection means, and the input image sample inspection Collation determination means for calculating the similarity in the partial space with the feature vector at the sample point on the input image detected by the output means, and comparing the input image with the registered image to confirm the identity. It is characterized by having.
[0013]
According to a second aspect of the present invention, in the image recognition apparatus according to the first aspect, the partial space creating means uses a correlation between a feature vector in the image in the image set and a feature vector in the average image. First detection means for detecting a sample point on the image in the image set corresponding to the sample point set on the average image, and an image in the image set detected by the first detection means Temporary subspace creation means for creating a partial space of a temporary feature vector using the feature vector at the upper sample point, and the average image based on the distance to the partial space of the feature vector in the image in the image set Second detection means for detecting a sample point on the image in the image set corresponding to the sample point set above, and the second detection means detected by the second detection means Characterized in that a subspace modifying means for modifying the partial space created by the provisional partial space creating means using the feature vectors at the sample point on the image of the image in the set.
[0014]
According to a third aspect of the present invention, in the image recognition apparatus according to the second aspect of the present invention, the provisional partial space creating means prepares an image pair of the person from the image set, and the first detecting means First principal difference subspace calculating means for calculating a principal difference subspace by calculating a difference between feature vectors at the sample points of the image pair corresponding to the detected sample point set on the average image; Prepare another person's image pair from the image set, and calculate the difference between the feature vectors at the sample points of the image pair corresponding to the sample points set on the average image detected by the first detection means. A first other person difference subspace calculating means for calculating the other person difference subspace, and calculating a crossing portion between the complement space of the principal difference subspace and the other person subspace, and setting the crossing portion to the subspace. Characterized by comprising a first recognition moiety space creating means for.
[0015]
According to a fourth aspect of the present invention, in the image recognition apparatus according to the second aspect of the present invention, the partial space correction means prepares an image pair of the user from the image set and detects the image pair by the second detection means. Second principal difference subspace calculating means for calculating a principal difference subspace by calculating a difference between feature vectors at the sample points of the image pair corresponding to the sample points set on the averaged image, An image pair of another person is prepared from the image set, and a difference between feature vectors at the sample points of the image pair corresponding to the sample points set on the average image detected by the second detection unit is obtained. A second person difference subspace calculating means for calculating a person difference subspace, and calculating an intersection between the complement space of the person difference subspace and the person difference subspace, and making the intersection a subspace. Second recognition subspace creation means, the second detection means, the second personal difference subspace calculation means, and the second until the partial space created by the second recognition subspace creation means converges. And a recognition subspace update means for updating the partial space by repeatedly using two different person subspace calculation means and the second recognition subspace creation means.
[0016]
An image recognition apparatus according to a fifth aspect of the present invention is the image recognition apparatus according to any one of the first to fourth aspects, wherein the registered image sample point detecting means is configured to register the registration points corresponding to the sample points on the average image. A sample point on the image is detected based on a distance between a feature vector in the registered image and the partial space.
[0017]
An image recognition apparatus according to a sixth aspect of the present invention is the image recognition apparatus according to any one of the first to fifth aspects, wherein the input image sample point detecting means corresponds to the input corresponding to the sample point on the registered image. A sample point on the image is detected based on a correlation in the partial space between the feature vector in the input image and the feature vector in the registered image.
[0018]
An image recognition apparatus according to a seventh aspect of the present invention is the image recognition apparatus according to any one of the first to sixth aspects, wherein the collation determining means corresponds to the sample points on the registered image and the input image. The similarity is calculated based on the correlation of the feature vector in the partial space, and the input image and the registered image are collated to confirm the identity.
[0019]
According to an eighth aspect of the present invention, there is provided an image recognition method including a local feature at a sample point between an input image obtained by capturing a predetermined part of a person's body and a plurality of registered images of the person and others relating to the part. In the image recognition method for verifying the identity by comparing the images by comparing the average image creation step of creating an average image from a plurality of images in the image set relating to the part, and the sample points set on the average image A subspace creating step for detecting a sample point on each image in the image set corresponding to the image and creating a subspace of a feature vector at the sample point; and on the registered image corresponding to the sample point on the average image A registered image sample point detecting step for detecting a sample point of the registered image using the partial space, and a sample on the input image corresponding to the sample point on the registered image An input image sample point detecting step for detecting points using the subspace, a feature vector at a sample point on the registered image detected by the registered image sample point detecting step, and an input image sample point detecting step And a collation determining step of calculating the similarity in the partial space with the feature vector at the sample point on the input image and collating the input image with the registered image to perform identity verification. Features.
[0020]
Further, the program according to the invention of claim 9 compares local features at sample points between an input image obtained by capturing a predetermined part of the body of the person and a plurality of registered images of the person and others regarding the part. A program for causing a computer to execute an image recognition method for verifying an identity by comparing images, and an average image creating step for creating an average image from a plurality of images in an image set related to the region; Detecting a sample point on each image in the image set corresponding to the set sample point and generating a subspace of a feature vector at the sample point; and a sample point on the average image A registered image sample point detecting step for detecting a corresponding sample point on the registered image using the partial space; and a sample on the registered image. An input image sample point detecting step for detecting a sample point on the input image corresponding to the input image using the partial space, a feature vector at the sample point on the registered image detected by the registered image sample point detecting step, and the Collation for calculating the similarity in the partial space with the feature vector at the sample point on the input image detected by the input image sample point detection step and collating the input image with the registered image for identity verification The determination step is caused to be executed by a computer.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
Exemplary embodiments of an image recognition apparatus, an image recognition method, and a program for causing a computer to execute the method according to the present invention will be described below in detail with reference to the accompanying drawings. Hereinafter, a case of face image recognition using a face image will be described.
[0022]
First, the concept of the image recognition process according to the present embodiment will be described. FIG. 1 is an explanatory diagram for explaining the concept of image recognition processing according to the present embodiment.
[0023]
As shown in FIG. 1, the face image recognition process according to the present embodiment is composed of two major processes, an offline learning process and a collation process, and knowledge 13 in which the facial image features are learned by the learning process. It is comprised so that a collation process may be performed using.
[0024]
First, a processing procedure of offline learning processing will be described. First, the positions of both eyes, nose, and mouth are designated for each image in the face image set 10. However, the positions of both eyes should be specified at least here. Then, normalization is performed so that the positions of both eyes, nose, and mouth of each face image overlap, and the average face image 11 is created by averaging the gray value of each pixel.
[0025]
Here, the “image set” is a collection of images related to a predetermined part of the body of a plurality of persons, and a plurality of different persons also differ depending on differences in facial expressions, palm open / close, imaging angle, etc. The person's own image is included. This image set need not be identical to the set of registered images.
[0026]
Next, an arbitrary number of sample points are designated on the created average face image 11. FIG. 1 shows an average face image example 12 in which 100 sample points are designated on the average face image 11. Then, using the correlation between the feature vector of the average face image 11 and the feature vector of each image in the face image set 10, sample points on each image in the face image set 10 corresponding to the sample points on the average face image 11 are used. Is detected. Then, on the detected sample points, a feature vector partial space in each image in the face image set 10 is created using a principal component analysis technique. This subspace is provisional and will be modified later.
[0027]
Here, the “feature vector” represents a collection of feature quantities such as a gray value of a pixel and a change rate of the gray value in a local region around the sample point as one vector. If a digital face image composed of a collection of a plurality of pixels each having a feature amount is given, it can be easily created.
[0028]
Here, the concept of the partial space used in the present embodiment will be described. FIG. 2 is an explanatory diagram for explaining the concept of the partial space according to the present embodiment. In the feature vector space 20, the person's variation subspace 22 is composed of differences in feature vectors of a plurality of individual image pairs depending on differences in facial expressions, shooting angles, and the like. It is comprised from the difference of the feature vector of an image pair. The target subspace 23 can be derived from the intersection of the person's own subspace 22 and the other person's subspace 21. Hereinafter, the partial space configured in this way is referred to as a “recognition partial space”.
[0029]
Returning to the description of FIG. 1, the provisional recognition subspace 23 of the feature vector created using the images in the face image set 10 is transferred to the feature vector recognition subspace 23 in each image in the face image set 10. It is corrected based on the distance.
[0030]
Specifically, the distance to the recognition subspace 23 of the feature vector in each image in the face image set 10 is calculated, and a sample point on each image that decreases the distance is detected. When a sample point is detected, a new recognition subspace 23 is created based on the feature vector in each image at that sample point, and the distance of the feature vector to the recognition subspace 23 until the created recognition subspace 23 converges. The recognition point space 23 is corrected by repeating the sample point detection processing based on the above and the partial space creation processing based on the feature vector at the sample point of each image in the face image set 10. The obtained recognition subspace 23 is used at the time of image collation as knowledge 13 obtained by learning the features of the face image.
[0031]
Here, the concept of detection of sample points of an image using the distance of the feature vector to the recognition subspace 23 used in the present embodiment will be described. FIG. 3 is an explanatory diagram for explaining the concept of sample point detection on a face image using the distance of the feature vector to the recognition subspace 23 according to the present embodiment.
[0032]
In FIG. 3, as the distance 31 of the feature vector 30 of the face image to the recognition subspace 23 is smaller, the feature corresponding to the facial feature extracted in the recognition subspace 23 appears more frequently on the face image. Means that That is, by selecting a sample point so that this distance becomes small, a sample point corresponding to the sample point on the average face image 11 can be set on the image.
[0033]
Returning to the description of FIG. 1, the processing procedure of the collation processing will be described next. First, the distance to the recognition subspace 23 of the feature vector in the registered image 14 is calculated, and sample points on the registered image that reduce the distance are detected. FIG. 1 shows an example 15 of a registered image in which sample points are detected on the registered image 14.
[0034]
As a method for selecting the registered image 14 to be collated with the input image 16 from a set of a plurality of registered images, a method for selecting the registered images 14 one after another from the set of the plurality of registered images, or a method for inputting by the person himself / herself. For example, there is a method of specifying and selecting the registered image 14 associated with the ID number.
[0035]
Here, “registered image” means a digital face image composed of a collection of a plurality of pixels each having a gray value, and a feature created in a local region around the pixels in the plurality of pixels of the digital face image This refers to feature vector information consisting of vectors. However, if at least the feature vector information is included in the registered image, it is possible to perform face image matching processing.
[0036]
For the input image 16, sample points corresponding to the detected sample points on the registered image 14 are detected using the correlation between the feature vectors of the input image 16 and the registered image 14 in the recognition subspace 23. FIG. 1 shows an input image example 17 in which sample points are detected on the input image 16.
[0037]
FIG. 4 is an explanatory diagram for explaining the concept of sample point detection on the input image 16 according to the present embodiment. As shown in FIG. 4, the feature vector of the input image 16 and the feature vector of the registered image 14 are projected onto the recognition subspace 23, and the feature vector 40 of the projected input image and the feature vector 41 of the projected registered image are projected. The correlation value between them is calculated as the cosine of the angle θ between the projected feature vectors, that is, cos θ. A larger value means that the features of both images at the sample point are similar.
[0038]
Returning to the description of FIG. 1, when the corresponding sample points on the registered image 14 and the input image 16 are detected, the feature vector of the registered image 14 projected into the recognition subspace 23 at the sample point and the input are next input. A correlation value between both feature vectors is calculated in the recognition subspace 23 using the feature vector of the image 16. This process is performed for all sample points. And the average value of the correlation value with respect to the image pair of the input image 16 and the registration image 14 is calculated, and it is set as similarity. In FIG. 1, in Example 12 in which 100 sample points are set on the average face image 11, an average value (similarity) of 100 correlation values is calculated. Then, based on the similarity, the person and others are identified.
[0039]
Next, the configuration of the face image recognition apparatus according to this embodiment will be described. FIG. 5 is a functional block diagram showing the configuration of the face image recognition apparatus according to the present embodiment.
[0040]
As shown in FIG. 5, the face image recognition device 50 according to the present embodiment includes a face image input receiving unit 51, a normalization processing unit 52, a face image learning unit 53, a registered image storage unit 54, and a registration. An image sample point detection unit 55, an input image sample point detection unit 56, a similarity calculation unit 57, and a collation determination unit 58 are provided.
[0041]
The face image input reception unit 51 is a reception unit for taking a learning face image set, an input face image, and a registration face image into the apparatus, and outputs the received image to the normalization processing unit 52.
[0042]
The normalization processing unit 52 acquires an image from the face image input receiving unit 51, normalizes the image to match the size of the face of each image, and the received image is an image of a learning face image set. If there is a registered face image, the normalized image is output to the registered image storage unit 54, and if it is an input face image, the normalized image is output to the input image sample point detecting unit 56.
[0043]
The face image learning unit 53 acquires a normalized face image set from the normalization processing unit 52, creates an average face image 11 using the face image set 10, and uses the average face image 11 to generate a feature vector. The recognition subspace 23 is created.
[0044]
The registration image storage unit 54 acquires and stores the registration image 14 normalized by the normalization processing unit 52 when the registration face image is received by the face image input reception unit 51.
[0045]
The registered image sample point detection unit 55 reads the recognition subspace 23 from the face image learning unit 53 and performs registration image 14 from the registered image storage unit 54 when collation determination is performed between the registered image 14 and the input image 16. And the sample point on the registered image 14 is detected using the distance of the feature vector in the registered image 14 to the recognition partial space 23.
[0046]
The input image sample point detection unit 56 reads the recognition subspace 23 from the face image learning unit 53 and performs the input image 16 from the normalization processing unit 52 when collation determination is performed between the registered image 14 and the input image 16. And the sample point information on the registered image 14 and the registered image 14 is read from the registered image sample point detection unit 55, and the correlation of the feature vectors of the registered image 14 and the input image 16 in the recognition subspace 23 is calculated. Sample points on the input image 16 are detected.
[0047]
The similarity calculating unit 57 recognizes the sample point information of the registered image 14 and the input image 16, and the recognition subspace 23 of the feature vector of the registered image 14 and the input image 16 at each sample point calculated by the input image sample point detecting unit 56. The correlation value data is read, and the similarity between the registered image 14 and the input image 16 is calculated.
[0048]
The collation determining unit 58 reads the similarity between the registered image 14 and the input image 16 obtained by the similarity calculating unit 57, performs collation determination between the registered image 14 and the input image 16 based on the similarity, and performs a collation result. Is output.
[0049]
Hereinafter, the image recognition processing according to the present embodiment shown in FIG. 1 will be described more specifically. FIG. 6 is a flowchart of normalization processing for normalizing face image data according to the present embodiment. This processing is performed by the normalization processing unit 52. Here, a case where the image received by the face image input receiving unit 51 is a learning face image set is shown.
[0050]
First, an average face image 11 is provisionally provided, and provisional positions of both eyes are set on the image (step S601). Next, the face image data in the face image set 10 is read through the face image receiving unit 51 (step S602), and the positions of both eyes, nose and mouth of each read face image data are designated (step S603). However, the positions of both eyes should be specified at least here.
[0051]
After the positions of both eyes, nose and mouth are designated, affine transformation is performed so that the positions of both eyes of each face image data overlap with the positions of both eyes of the average face image 11 (step S604). The normal positions of both eyes, nose and mouth are calculated (step S605). Then, the average value of the calculated normal positions of both eyes, nose and mouth is calculated (step S606), and the positions of both eyes, nose and mouth of the average face image 11 are set (step S607).
[0052]
When the positions of both eyes, nose and mouth of the average face image 11 are set, affine transformation is performed so that the positions of both eyes, nose and mouth of each face image data overlap with the positions of both eyes, nose and mouth of the average face image 11. (Step S608), a normalized image of each face image in the image set 10 is created (Step S609).
[0053]
Although the case where the image received by the face image input receiving unit 51 is the learning face image set 10 is shown here, the learning process is completed and the received image is the input face image or the registration face. If it is an image, the positions of both eyes, nose and mouth of the average face image 11 have already been set during learning, so the face image data is read (step S602), and both eyes and nose of the read face image data are read. The position of the mouth is designated (step S603), and affine transformation is performed so that the positions of both eyes, nose and mouth of the face image data overlap with the positions of both eyes, nose and mouth of the average face image 11 (step S608). It is only necessary to create a normalized image of the image (step S609).
[0054]
Next, the process of creating the feature vector recognition subspace 23 from the face image set 10 in the face image learning unit 53 will be described. FIG. 7 is a flowchart of a recognition subspace creation process for creating a feature vector recognition subspace 23 from the face image set 10 according to the present embodiment.
[0055]
First, N face image data are read from the learning face image set 10 (step S701). The average of the shade values of the corresponding pixels of the read N face image data is calculated (step S702), and the average face image 11 is created (step S703). Then, M sample points are set on the created average face image 11 (step S704).
[0056]
Similarly, M sample points corresponding to the sample points on the average face image 11 are set on each face image data in the N face image sets 10 (step S705), and the corresponding M sample points are set. Above, the correlation between the average face image 11 and the feature vectors of the N face image data is calculated (step S706). It is checked whether or not the correlation between the feature vectors is larger than a certain threshold (step S707). If the correlation is large, samples corresponding to M sample points of the average face image 11 on each of N face image data. A point is determined (step S708). If the correlation is small, M sample points are set again on each of the N face image data (step S705), and these processes (steps S705 to S707) are repeated until a sample point with a large correlation is found. .
[0057]
Then, after confirming sample points corresponding to M sample points of the average face image 11 on each of N face image data (step S708), each of M sample points on each of N face image data. M feature vector recognition subspaces 23 are created (step S709).
[0058]
Next, a specific creation process for creating the recognition subspace 23 at the sample points in the process of creating the recognition subspace 23 in FIG. 7 (step S709) will be described. FIG. 8 is a flowchart of a recognition subspace creation process for creating a recognition subspace 23 at sample points according to the present embodiment.
[0059]
First, Ns sets of personal image data pairs are selected from the face image data in the N learning face image sets 10 (step S801). The personal image data pair is composed of a plurality of personal images that differ depending on differences in facial expressions, photographing angles, and the like. Then, the difference between the feature vectors at the corresponding sample points of the data pair is calculated (step S802), and the principal difference subspace Ss is calculated based on the difference (step S803).
[0060]
Next, a data pair of images of other people is selected from the face image data in the N learning face image sets 10 (step S804), and a feature vector difference at a corresponding sample point of the data pair is calculated. (Step S805), and the other person difference subspace So is calculated based on the result (Step S806).
[0061]
When the personal difference subspace Ss and the different person subspace Ss are calculated, the complementary space Ss of the personal difference subspace Ss is calculated.^CAnd S between other people's difference subspace Sо, that is,
S = So ∩ Ss^C
Is calculated as a recognition subspace 23 (step S807).
[0062]
Next, a specific process for creating a subspace based on a feature vector at a sample point in the process of creating the principal difference subspace Ss and the other person difference subspace Sо in FIG. 8 (steps S803 and S806). explain. FIG. 9 is a flowchart of a partial space creation process for creating a partial space based on the feature vector according to the present embodiment.
[0063]
First, R-dimensional N feature vectors Xi (1 ≦ i ≦ N) are input (step S901). And the correlation matrix M of R rows and R columns is
M (k, l) = 1 / N Σ Xi (k) Xi (l)
Is calculated using the above formula (step S902). Here, 1 ≦ k and l ≦ R.
[0064]
When the correlation matrix M is obtained, R eigenvalues are calculated using the correlation matrix M (step S903), and r (r ≦ R) eigenvalues are extracted from the larger one (step S904) to correspond to the r eigenvalues. A space spanned by the eigenvector is calculated to obtain an r-dimensional subspace (step S905). This is mathematically the same as the statistical method called principal component analysis.
[0065]
Next, the complementary space Ss of the personal difference subspace Ss in FIG.^CIn the process of creating the recognition subspace 23 by calculating the intersection of the person difference subspace So and the other person difference subspace So (step S807), the specific subspace 23 is calculated from the person difference subspace Ss and the person difference subspace So. Processing will be described. FIG. 10 is a flowchart of a recognition subspace calculation process for calculating the recognition subspace 23 according to the present embodiment.
[0066]
First, the projection matrix Ps of the personal difference subspace Ss is calculated (step S1001), and then the projection matrix Po of the different person subspace So is calculated (step S1002). Then, set a constant α (0 ≦ α ≦ 1)
P = α Pо + (1 − α) (I − Ps)
Thus, the matrix P is calculated (step S1003). Here, I is a unit matrix. α is a parameter for adjusting the weight between the different person difference subspace and the complement of the person difference subspace, and can be arbitrarily specified. Hereinafter, this matrix P is referred to as a mixing matrix.
[0067]
Then, an eigenvalue of the mixing matrix is calculated (step S1004), and a threshold for the eigenvalue is set (step S1005). The possible range of eigenvalues of the mixing matrix is 0 or more and 1 or less, and the threshold is set to a value close to 1 or 1 or less. Then, eigenvalues greater than or equal to the threshold are extracted, a space spanned by eigenvectors for the eigenvalues is calculated, and set as a recognition subspace 23 (step S1006).
[0068]
Next, in the face image learning unit 53, specific processing for learning facial features while correcting the recognition subspace 23 created by the processing shown in FIG. 7 using the learning face image set 10. Will be described. FIG. 11 is a flowchart of recognition subspace correction processing for correcting the feature vector recognition subspace 23 from the face image set 10 according to the present embodiment.
[0069]
First, the maximum number of iterations of the convergence calculation of the recognition subspace 23 is set (step S1101), and M sample points on the average face image 11 are placed on each of the N face images in the learning face image set 10. Corresponding M sample points are set (step S1102). Then, the distances to the M recognition subspaces 23 corresponding to the M feature vectors in the N face images are calculated (step S1103).
[0070]
Then, it is checked whether or not the distance to the recognition subspace 23 is smaller than a certain threshold value (step S1104). If the distance is small, M samples on the average face image 11 are included in each of N face image data. A sample point corresponding to the point is determined (step S1105). If the distance is large, M sample points are set again on each of the N face image data (step S1102), and these processes (steps S1102 to S1104) are repeated until a sample point with a small distance is found. .
[0071]
When sample points corresponding to M sample points on the average face image 11 are determined on the N face image data (step S1105), the sample points at the M sample points on the N face image data are determined. M feature vector recognition subspaces 23 are created (step S1106). Then, it is checked whether or not the recognition subspace 23 has converged (step S1107). When the recognition subspace 23 has converged, the correction processing of the recognition subspace 23 is finished assuming that the learning has ended. If not converged, it is next checked whether or not the number of iterations for convergence calculation is smaller than the set maximum number of iterations (step S1108). In this case, M sample points are set again on each of the N face image data (step S1102), and these processes (steps S1102 to S1108) are repeated until the recognition subspace 23 converges.
[0072]
Next, a specific procedure of the process (step S1107) for checking whether or not the recognition subspace 23 has converged shown in FIG. 11 will be described. FIG. 12 is a flowchart of a recognition subspace convergence determination process for determining convergence of the recognition subspace 23 according to the present embodiment.
[0073]
First, a canonical angle threshold is set for the two recognition subspaces 23 before and after correction (step S1201), and the canonical angle is calculated (step S1202). Then, it is checked whether or not the smallest canonical angle among the plurality of canonical angles is smaller than the threshold value (step S1203). If smaller, it is determined that the recognition subspace 23 has converged (step S1204). Determines that the recognition subspace 23 has not converged (step S1205).
[0074]
Next, the processing procedure of the collation determination process for collating the registered image 14 and the input image 16 to confirm the identity will be described. FIG. 13 is a flowchart showing a processing procedure of collation determination processing according to the present embodiment.
[0075]
First, the registered image sample point detection unit 55 reads the registered image 14 stored in the registered image storage unit 54 and the data of the recognition subspace 23 created by the face image learning unit 53, and reads the average face on the registered image 14. M sample points corresponding to the sample points on the image 11 are set (step S1301). Then, the distance of the feature vector of the registered image 14 to the recognition subspace 23 is calculated on each corresponding sample point (step S1302). The concept of this distance has already been explained in FIG.
[0076]
Thereafter, it is checked whether or not the distance to the recognition subspace 23 is smaller than a certain threshold value (step S1303). If the distance is small, M sample points of the average face image 11 on each of the N face image data. A sample point corresponding to is determined (step S1304). If the distance is not small, M sample points are reset on each of the N face image data (step S1301), and these processes (steps S1301 to S1303) are performed until a sample point with a small distance is found. repeat. These processes (steps S1301 to S1304) are performed by the registered image sample point detection unit 55.
[0077]
Next, the input image sample point detection unit 56 includes the input image 16 normalized by the normalization processing unit 52, the recognition subspace data 23 created by the face image learning unit 53, and the registered image sample point detection unit 55. The detected sample point data and the normalized registered image 14 are read, and M sample points corresponding to the sample points on the registered image 14 are set on the input image 16 (step S1305). Then, the correlation of the feature vectors of the registered image 14 and the input image 16 on the corresponding sample points in each recognition subspace 23 is calculated (step S1306).
[0078]
Thereafter, it is checked whether or not the correlation between the feature vectors of the registered image 14 and the input image 16 in the recognition subspace 23 is larger than a certain threshold (step S1307). If the correlation is large, the registered image is displayed on the input image 16. The sample points corresponding to the M sample points on 14 are determined (step S1308). If the correlation is small, M sample points are set again on the input image 16 (step S1305), and these processes (steps S1305 to S1307) are repeated until a sample point whose correlation is greater than the threshold is found. These processes (steps S1305 to S1308) are performed by the input image sample point detection unit 56.
[0079]
The concept of the correlation in the recognition subspace 23 has already been described with reference to FIG. 4, and specifically, between the feature vectors of the registered image 14 and the input image 16 projected onto the recognition subspace 23. The cosine of the angle θ formed, that is, cos θ.
[0080]
When the corresponding sample points of the registered image 14 and the input image 16 are determined (step S1308), the similarity calculation unit 57 inputs the correlation value data of the registered image 14 and the input image 16 between the determined sample points. Reading from the image sample point detection unit 56, the similarity between the registered image 14 and the input image 16 is calculated by obtaining the average value of the correlation values at each sample point (step S1309). Then, the collation determination unit 58 performs collation determination between the registered image 14 and the input image 16 based on the degree of similarity (step S1310), and outputs a determination result as to whether or not the person corresponds to the registered image 14. (Step S1311).
[0081]
Next, the hardware configuration of the face image recognition apparatus according to this embodiment will be described. FIG. 14 is a block diagram showing a system configuration of the face image recognition apparatus according to the present embodiment.
[0082]
As shown in FIG. 14, the face image recognition device 50 includes an interface unit 64, a calculation control unit 65, a main storage unit 66, and an auxiliary storage unit 67, and includes an input device 68 and a display device 69. This is a connected stand-alone configuration using a computer.
[0083]
In addition, a CCD camera 60 for capturing an input image is connected to the face image recognition device 50 via an interface 64, and the CCD camera 60 includes an image input unit 61, an image memory 62, and an interface unit 63. It is configured.
[0084]
First, the configuration of the CCD camera 60 will be specifically described. The image input unit 61 collects light from the face that is the subject with a lens, converts the face image into an electrical signal using a CCD (Charge Coupled Device), converts the face image into digital data, and converts the face image data into an image. Record in the memory 62.
[0085]
The image memory 62 is used as a buffer memory of the image input unit 61, and temporarily stores the face image data when the face image recognition device 50 cannot accept the input of the face image. When the face image recognition device 50 is ready to accept an input image, the face image data is output to the face image recognition device 50 through the interface unit 63.
[0086]
Next, the configuration of the face image recognition device 50 will be specifically described. The interface unit 64 receives an input image from the CCD camera 60, receives data from the input device 68, and transfers data to the display device 69 under the control of the arithmetic control unit 65.
[0087]
The input device 68 is composed of a keyboard, a mouse, a touch panel, etc., and when executing the face image recognition program, when specifying the positions of both eyes, nose, and mouth with respect to the face image (step S603), a sample is formed on the created average face. This is used when a point is designated (step S704).
[0088]
The display device 69 is a display monitor, and when executing the face image recognition program, display of sample points on the average face image 11, display of sample points on the registered image 14, display of sample points on the input image 16, or This is used for displaying the collation determination result (step S1311).
[0089]
The auxiliary storage unit 67 is, for example, a CD drive device that can read data from a floppy (registered trademark) disk drive device (FDD), a hard disk drive device (HDD), a CD-ROM, a CD-R, a CD-RW, etc. A DVD drive device that can read data from a DVD-ROM, DVD-R, DVD-RW, DVD-RAM, or the like.
[0090]
The face image recognition program executed by the face image recognition apparatus 50 according to the present embodiment is recorded and provided as a file in an executable format on an FD, CD-ROM, DVD-ROM, or the like. The program is read and executed by a floppy (registered trademark) disk drive device, a CD drive device, a DVD drive, or the like.
[0091]
Further, the face image data in the learning face image set 10 and the face image data for registration are also provided by FD, CD-ROM, DVD-ROM or the like, and are a floppy (registered trademark) disk drive device, CD drive device or DVD. The face image data for learning is read out by the drive and stored and processed in the face image learning unit 53, and the face image data for registration is stored in the registered image storage unit 54. The auxiliary storage unit 67 constitutes a storage unit of the present invention.
[0092]
The auxiliary storage unit 67 does not need to be directly connected to the face image recognition device 50, and the auxiliary storage unit 67 may be present on the network. For example, a face image server may be installed on the Internet or a LAN to store learning face image data and registration face image data, and download them as necessary. In this case, the face image recognition device 50 needs to be further provided with a communication unit such as a modem or a LAN board.
[0093]
Further, the face image recognition apparatus 50 of the present embodiment is provided with an arithmetic control unit 65 such as a CPU for controlling the entire system, and a main storage unit 66 composed of storage media such as a RAM and a ROM. A boot program and the like are stored in advance in the ROM, and a part of an OS (operation system) read from the HD, a face image recognition program, and the like are stored in the RAM, and the arithmetic control unit 65 executes these programs. Further, the RAM stores various face image data, sample point data, calculation results, and the like read out when the face image recognition program is executed.
[0094]
As described above, in the present embodiment, first, an average face image 11 used to designate sample points from the face image set 10 is created, and the face image set 10 corresponding to the sample points on the average face image 11 is created. Sample points on each face image are detected to create a feature vector recognition subspace 23, and the sample points on the registered image 14 corresponding to the sample points on the average face image 11 using the recognition subspace 23. Are detected, sample points on the input image 16 corresponding to the sample points on the registered image 14 are detected using the recognition subspace 23 and the registered image 14, and the recognized partial space of the detected registered image 14 and input image 16 is detected. 23, the similarity of the feature vectors in the image 23 is calculated and the registered image 14 and the input image 16 are collated. Therefore, the sample points need only be specified on the average face image 11. It is possible to reduce the work load on the.
[0095]
In addition, a partial space is used for collation judgment, and the partial space is not easily affected by fluctuations of the person due to differences in facial expressions, shooting angles, etc. Since it is configured as the recognition subspace 23, it is possible to enhance the discrimination ability between the person image and the other person image.
[0096]
In the present embodiment, the case where the recognition target image of the image recognition apparatus is a face image has been shown. However, the present invention is not limited to this, and a predetermined body including a mark is included. It can also be applied to sites. For example, in the palm print example, the hand muscles can be used as the one corresponding to the eyes and nose of the face.
[0097]
That is, in the case of a face image, the positions of both eyes, nose, and mouth are specified when creating the average face image 11 (step S603), but in the case of a palm print image, the position of the hand muscles may be specified instead. Further, as the feature vector, a feature amount such as the direction of the raised line of the palm print, the curvature, etc. may be used. Since the other processes are the same as those of the face image, the palm print image recognition apparatus can be easily realized.
[0098]
【The invention's effect】
As described above, according to the invention of claim 1, an average image is created from a plurality of image sets related to a predetermined part of the body, and images in the image set corresponding to the sample points set on the average image The upper sample point is detected to create a feature vector subspace at the sample point, and the sample point on the registered image corresponding to the sample point on the set average image is detected using the created subspace. The sample points on the input image corresponding to the detected sample points on the registered image are detected using the created subspace, and the registered image detected in the created subspace and the sample points on the input image Therefore, it is not necessary to specify sample points on each registered image, and the sample points are specified only on the average image. Because it is sufficient, to reduce the burden of the work according to the specification of the sample points, an effect that sample point storing stores capacity can be reduced as the image processing device can be obtained.
[0099]
In addition, since the feature of each individual image is expressed using the partial space of the feature vector, the features common to all people are removed, the individual features are emphasized, and the features of the image can be efficiently extracted as knowledge. There is an effect that an image processing apparatus capable of improving the discrimination ability between the person image and the other person image is obtained.
[0100]
According to the second aspect of the present invention, on the image in the image set corresponding to the sample point set on the average image using the correlation between the feature vector in the image in the image set and the feature vector in the average image. A sub-space of the feature vector in the image in the image set is created by using the feature vector at the sample point on the image in the detected image set A sample point on the image in the image set corresponding to the sample point set on the average image based on the distance to the space is detected, and a feature vector at the sample point on the image in the detected image set is detected. Since the partial space of the feature vector created by the provisional partial space creating means is modified, the sample vector on the average image is created when creating the partial space of the feature vector. The sample point on the image of the image set corresponding to Le point detected efficiently, an effect that the image recognition apparatus capable of creating a subspace can be obtained.
[0101]
According to the invention of claim 3, an image pair of the person is prepared from the image set, and the difference between the feature vectors at the sample points of the image pair corresponding to the sample points set on the average image is obtained. The difference between the feature vectors at the sample points of the image pair corresponding to the sample points set on the average image is obtained by calculating the difference subspace of the person and preparing another person's image pair from the image set. Since the subspace is calculated, and the intersection of the calculated difference space of the personal difference subspace and the subspace of the other person is calculated and the intersection is defined as the subspace, the fluctuation of the personal image is excluded. It is resistant to noise caused by fluctuations in the person's image due to differences in facial expressions, palm opening, shooting angle, etc., and fluctuations between the person and others are emphasized, so knowledge can be efficiently extracted into the subspace. ,Book Image and others image and discriminating ability image recognition apparatus which can improve an effect that can be obtained.
[0102]
According to the invention of claim 4, an image pair of the person is prepared from the image set, and the difference between the feature vectors at the sample points of the image pair corresponding to the sample points set on the average image is obtained. The difference between the feature vectors at the sample points of the image pair corresponding to the sample points set on the average image is obtained by calculating the difference subspace of the person and preparing another person's image pair from the image set. Calculating the subspace, calculating the intersection of the calculated subspace of the principal difference and the subspace of the other person and making the intersection as a subspace, until the subspace created there converges The subspace is updated by repeatedly using the detecting means, the second person difference subspace calculating means, the second other person difference subspace calculating means, and the second recognition subspace creating means. Part It is possible to create a subspace in which the feature of the image set is extracted with higher accuracy than the provisional subspace created by the space creation means, and when identifying the person image and the other person image using the subspace There is an effect that an image recognition apparatus capable of improving the discrimination ability is obtained.
[0103]
Further, according to the invention of claim 5, since the sample point on the registered image corresponding to the sample point on the average image is detected based on the distance between the feature vector and the partial space in the registered image, Even if there is a difference in facial expression, palm opening, shooting angle, etc. between the average image and the registered image, it is possible to accurately detect sample points on the registered image corresponding to the sample points of the average image An effect is obtained that a simple image recognition apparatus can be obtained.
[0104]
According to the invention of claim 6, the sample points on the input image corresponding to the detected sample points on the registered image are obtained by subtracting the feature vector in the input image and the feature vector in the registered image in the subspace. Since the detection is based on the correlation, the input image is collated with the registered image even if there is a difference in facial expression, palm opening, shooting angle, etc. between the input image and the registered image. There is an effect that an image recognition apparatus capable of accurately detecting the upper sample point is obtained.
[0105]
According to the invention of claim 7, the similarity is calculated based on the correlation in the partial space of the feature vector between the detected registered image and the sample points on the input image, and the input image and the registered image are obtained. Because verification is performed and identity verification is performed, even if there is a difference in facial expression, open / closed palm, shooting angle, etc. between the input image and the registered image, it is possible to accurately and efficiently perform verification determination. There exists an effect that an image recognition device is obtained.
[0106]
According to the invention of claim 8, an average image is created from a plurality of image sets relating to a predetermined part of the body, and the sample points on the image in the image set corresponding to the sample points set on the average image And a feature vector partial space at the sample point is created, and a sample point on the registered image corresponding to the set sample point on the average image is detected using the created partial space, The sample points on the input image corresponding to the sample points on the registered image are detected using the created subspace, and the registered image detected in the created subspace and the feature vectors at the sample points on the input image Since the similarity is calculated and the input image and the registered image are checked for identity verification, there is no need to specify a sample point on each registered image, and it is only necessary to specify a sample point on the average image. , To reduce the burden of the work according to the specification of the sample points, an effect that the image recognition method capable of reducing the storage capacity for storing sample points are obtained.
[0107]
In addition, since the feature of each individual image is expressed using the partial space of the feature vector, the features common to all people are removed, the individual features are emphasized, and the features of the image can be efficiently extracted as knowledge. There is an effect that an image recognition method capable of improving the discrimination ability between the person image and the other person image is obtained.
[0108]
According to the invention of claim 9, an average image is created from a plurality of image sets relating to a predetermined part of the body, and the sample points on the image in the image set corresponding to the sample points set on the average image And a feature vector partial space at the sample point is created, and a sample point on the registered image corresponding to the set sample point on the average image is detected using the created partial space, The sample points on the input image corresponding to the sample points on the registered image are detected using the created subspace, and the registered image detected in the created subspace and the feature vectors at the sample points on the input image Since the similarity is calculated and the input image and the registered image are checked for identity verification, there is no need to specify a sample point on each registered image, and it is only necessary to specify a sample point on the average image. , To reduce the burden of the work according to the specification of the sample points, an effect that can program to reduce the storage capacity for storing the sample points are obtained.
[0109]
In addition, since the feature of each individual image is expressed using the partial space of the feature vector, the features common to all people are removed, the individual features are emphasized, and the features of the image can be efficiently extracted as knowledge. There is an effect that it is possible to obtain a program capable of improving the discrimination ability between the person image and the other person image.
[Brief description of the drawings]
FIG. 1 is an explanatory diagram for explaining a concept of image recognition processing according to the present embodiment;
FIG. 2 is an explanatory diagram for explaining a concept of a partial space according to the present embodiment.
FIG. 3 is an explanatory diagram for explaining a concept of sample point detection on a face image using a distance from a feature vector to a recognition subspace according to the present embodiment;
FIG. 4 is an explanatory diagram for explaining a concept of sample point detection on an input image according to the present embodiment.
FIG. 5 is a functional block diagram showing a configuration of a face image recognition apparatus according to the present embodiment.
FIG. 6 is a flowchart of normalization processing for normalizing face image data according to the present embodiment.
FIG. 7 is a flowchart of recognition subspace creation processing for creating a feature vector recognition subspace from a face image set according to the present embodiment;
FIG. 8 is a flowchart of a recognition subspace creation process for creating a recognition subspace at sample points according to the present embodiment.
FIG. 9 is a flowchart of a partial space creation process for creating a partial space based on a feature vector according to the present embodiment.
FIG. 10 is a flowchart of a recognition subspace calculation process for calculating a recognition subspace according to the present embodiment.
FIG. 11 is a flowchart of a recognition subspace correction process for correcting a feature vector recognition subspace from a face image set according to the present embodiment;
FIG. 12 is a flowchart of recognition subspace convergence determination processing for determining the convergence of the recognition subspace according to the present embodiment.
FIG. 13 is a flowchart showing a processing procedure of collation determination processing according to the present embodiment.
FIG. 14 is a block diagram showing a system configuration of a face image recognition apparatus according to the present embodiment.
[Explanation of symbols]
10 Face image set
11 Average face image
12 Example of average face image with 100 sample points specified
13 Knowledge
14 Registration images
15 Example of registered image with sample points detected
16 input images
17 Example of input image with sample points detected
20 Feature vector space
21 Other people's variable subspace
22 Fluctuating subspace of the person
23 Target recognition subspace
30 Feature vector of face image
31 Distance of feature vector to recognition subspace
40 Projection of feature vector of input image to recognition subspace
41 Projection of feature vector of registered image to recognition subspace
50 Face image recognition device
51 Face image input reception part
52 Normalization processing unit
53 Face Image Learning Department
54 Registered image storage
55 Registered image sample point detector
56 Input image sample point detector
57 Similarity detection unit
58 Collation determination unit
61 Image input section
62 Image memory
63, 64 interface part
65 Arithmetic Control Unit (CPU)
66 Main memory (ROM, RAM)
67 Auxiliary storage
68 Input device
69 Display device

Claims (9)

An image for verifying an identity by comparing local features at sample points between an input image obtained by capturing a predetermined part of the person's body and a plurality of registered images of the person and others relating to the part. In the recognition device,
Average image creating means for creating an average image from a plurality of images in the image set relating to the part;
A subspace creating means for detecting a sample point on each image in the image set corresponding to the sample point set on the average image and creating a subspace of the feature vector at the sample point;
Registered image sample point detection means for detecting a sample point on the registered image corresponding to a sample point on the average image using the partial space;
Input image sample point detection means for detecting a sample point on the input image corresponding to a sample point on the registered image using the partial space;
Within the partial space between the feature vector at the sample point on the registered image detected by the registered image sample point detection means and the feature vector at the sample point on the input image detected by the input image sample point detection means An image recognition apparatus comprising: a collation determining unit that calculates the similarity of the input image and collates the input image with the registered image to perform identity verification. The partial space creating means uses the correlation between the feature vector in the image in the image set and the feature vector in the average image to set the image on the image in the image set corresponding to the sample point set on the average image. First detection means for detecting sample points;
Provisional subspace creation means for creating a subspace of a provisional feature vector using a feature vector at a sample point on an image in the image set detected by the first detection means;
Second detection means for detecting a sample point on the image in the image set corresponding to the sample point set on the average image based on a distance of the feature vector in the image in the image set to the partial space When,
Subspace correction means for correcting the subspace created by the provisional subspace creation means using a feature vector at a sample point on an image in the image set detected by the second detection means. The image recognition apparatus according to claim 1. The provisional subspace creating means prepares an image pair of the person from the image set, and the sample of the image pair corresponding to the sample point set on the average image detected by the first detecting means A first principal difference subspace calculating means for calculating a principal difference subspace by obtaining a difference between feature vectors at points;
Prepare another person's image pair from the image set, and calculate the difference between the feature vectors at the sample points of the image pair corresponding to the sample points set on the average image detected by the first detection means. A first other person difference subspace calculating means for calculating and calculating another person difference subspace;
3. A first recognition subspace creating means for calculating a crossing portion between the complementary space of the personal difference subspace and the different person subspace and using the crossing portion as a subspace. The image recognition apparatus described in 1. The subspace correction means prepares an image pair of the person from the image set, and the sample points of the image pair corresponding to the sample points set on the average image detected by the second detection means Second principal difference subspace calculating means for calculating a principal difference subspace by obtaining a difference between feature vectors in
Preparing an image pair of another person from the image set, and calculating a difference between the feature vectors at the sample points of the image pair corresponding to the sample points set on the average image detected by the second detection unit; Second other person difference subspace calculating means for calculating and calculating another person difference subspace;
A second recognizing subspace creating means for calculating a crossing part between the complementary space of the personal difference subspace and the different person subspace and using the crossing part as a subspace;
Until the partial space created by the second recognition subspace creating means converges, the second detecting means, the second personal difference subspace calculating means, the second different person difference subspace calculating means, The image recognition apparatus according to claim 2, further comprising a recognition subspace updating unit that updates the partial space by repeatedly using a second recognition subspace creation unit. The registered image sample point detecting means detects a sample point on the registered image corresponding to a sample point on the average image based on a distance between a feature vector in the registered image and the partial space. The image recognition apparatus as described in any one of Claims 1-4. The input image sample point detection unit is configured to determine a sample point on the input image corresponding to the sample point on the registered image based on a correlation in the partial space between a feature vector in the input image and a feature vector in the registered image. The image recognition apparatus according to claim 1, wherein the image recognition apparatus detects the image recognition apparatus. The collation determining means calculates similarity based on the correlation in the partial space of feature vectors at the corresponding sample points on the registered image and the input image, and calculates the input image and the registered image. The image recognition apparatus according to claim 1, wherein collation is performed to confirm the identity. An image for verifying an identity by comparing local features at sample points between an input image obtained by capturing a predetermined part of the person's body and a plurality of registered images of the person and others relating to the part. In the recognition method,
An average image creating step of creating an average image from a plurality of images in the image set relating to the part;
A subspace creating step of detecting a sample point on each image in the image set corresponding to the sample point set on the average image and creating a subspace of the feature vector at the sample point;
A registered image sample point detecting step of detecting, using the partial space, a sample point on the registered image corresponding to a sample point on the average image;
An input image sample point detection step of detecting a sample point on the input image corresponding to a sample point on the registered image using the partial space;
In the partial space between the feature vector at the sample point on the registered image detected by the registered image sample point detection step and the feature vector at the sample point on the input image detected by the input image sample point detection step An image recognition method comprising: a collation determination step of calculating a similarity level of the input image and collating the input image with the registered image to perform identity verification. An image for verifying an identity by comparing local features at sample points between an input image obtained by capturing a predetermined part of the person's body and a plurality of registered images of the person and others relating to the part. A program for causing a computer to execute a recognition method,
An average image creating step of creating an average image from a plurality of images in the image set relating to the part;
A subspace creating step of detecting a sample point on each image in the image set corresponding to the sample point set on the average image and creating a subspace of the feature vector at the sample point;
A registered image sample point detecting step of detecting, using the partial space, a sample point on the registered image corresponding to a sample point on the average image;
An input image sample point detection step of detecting a sample point on the input image corresponding to a sample point on the registered image using the partial space;
In the partial space between the feature vector at the sample point on the registered image detected by the registered image sample point detection step and the feature vector at the sample point on the input image detected by the input image sample point detection step A program for causing a computer to execute a collation determination step of calculating the similarity and collating the input image with the registered image to confirm the identity.

Images