Processing device, processing system, processing method, and recording medium

Abstract

This processing device includes a processing means that generates a new image for training a classifier for classifying images on the basis of an inter-eye distance of a face included in an original image and a shooting distance assumed for the new image to be generated from the original image.

Description

Processing device, processing system, processing method, and recording medium

[0001] The present disclosure relates to a processing device, a processing system, a processing method, and a recording medium.

[0002] Facial recognition technology is used in a wide range of fields as a type of biometric authentication. Patent Document 1 discloses, as a related technology, a technology relating to an imaging device that enables easy acquisition of an image captured from a desired position. Patent Document 2 discloses, as a related technology, a technology relating to an image processing device that can efficiently detect a face of a size required by a user. Patent Document 3 discloses, as a related technology, a technology relating to an image processing device that can acquire a natural-looking image even after performing blurring processing according to a distance map.

[0003] International Publication No. 2021 / 149509 Japanese Patent Application Laid-Open No. 2021-064043 Japanese Patent Application Laid-Open No. 2015-046678

[0004] Incidentally, face recognition technology sometimes uses a classifier to classify face images. The face images input to this classifier are not necessarily shot at a constant distance. In other words, the classifier is sometimes required to be able to classify face images shot at various distances. To allow the classifier to classify face images shot at various distances, the classifier needs to be trained (internal parameters of the classifier need to be determined) using face images shot at various distances. However, preparing such face images shot at various distances takes time and effort. Therefore, there is a need for technology that can generate new images from existing images to train a classifier.

[0005] One of the objectives of each aspect of the present disclosure is to provide a processing device, a processing system, a processing method, and a recording medium that can solve the above-mentioned problems.

[0006] According to one aspect of the present disclosure, a processing device includes a processing means for generating a new image for training a classifier that classifies images based on the eye distance of a face included in an original image and an assumed shooting distance for the new image generated from the original image.

[0007] According to another aspect of the present disclosure, a processing system includes the above-described processing device and a control device that performs predetermined control based on a result of image classification by the processing device.

[0008] According to another aspect of the present disclosure, a processing method generates a new image for training a classifier that classifies images based on the eye distance of a face included in an original image and an assumed shooting distance for the new image generated from the original image.

[0009] According to another aspect of the present disclosure, a recording medium stores a program for causing a computer to execute the following steps: generate a new image for training a classifier that classifies images, based on the eye distance of a face included in an original image and an assumed shooting distance for the new image to be generated from the original image.

[0010] According to aspects of the present disclosure, new images can be generated from existing images to train a classifier.

[0011] FIG. 1 is a diagram illustrating an example of a configuration of a processing system according to some embodiments of the present disclosure. FIG. 2 is a diagram illustrating an example of a configuration of a processing device according to some embodiments of the present disclosure. FIG. 3 is a diagram for explaining processing performed by a first reproduction unit according to some embodiments of the present disclosure. FIG. 4 is a diagram illustrating a correspondence relationship between eye-to-eye distance and shooting distance. FIG. 5 is a diagram illustrating an example of an original image according to some embodiments of the present disclosure. FIG. 6 is a diagram illustrating an example of an image obtained by adding noise to an image according to some embodiments of the present disclosure. FIG. 7 is a diagram illustrating an example of a configuration of a control device according to some embodiments of the present disclosure. FIG. 8 is a diagram illustrating a first example of a processing flow of a processing system according to some embodiments of the present disclosure. FIG. 9 is a diagram illustrating a second example of a processing flow of a processing system according to some embodiments of the present disclosure. FIG. 10 is a diagram illustrating an example of a configuration of a processing device according to another embodiment of the present disclosure. FIG. 11 is a diagram illustrating an example of a configuration of a processing device according to another embodiment of the present disclosure. FIG. 12 is a diagram illustrating an example of a processing flow of a processing device according to another embodiment of the present disclosure. FIG. 13 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment.

[0012] Hereinafter, an embodiment will be described in detail with reference to the drawings. <Embodiment> A processing system 1 according to an embodiment of the present disclosure is a system that can generate new images for training a classifier from existing images.

[0013] 1 is a diagram illustrating an example of the configuration of a processing system 1 according to some embodiments of the present disclosure. As shown in FIG. 1, the processing system 1 includes a processing device 10 and a control device 20.

[0014] 2 is a diagram illustrating an example of the configuration of a processing device 10 according to some embodiments of the present disclosure. The processing device 10 generates a new image from an existing image that can be treated as an image captured at a longer distance than the shooting distance at which the existing image was captured. The processing device 10 then trains a classifier 107 (described later) using the existing image and the generated new image. As shown in FIG. 2 , the processing device 10 includes a first input unit 101, a detector 102, a crop unit 103, a second input unit 104, a generation unit 105, a resizing unit 106, and a classifier 107.

[0015] The first input unit 101 inputs a plurality of original images to the detector 102. Each of the original images is an image in which the entire face is captured.

[0016] The detector 102 detects facial feature points in each of the original images. The detector 102 performs the detection using, for example, a cascade shape regression technique. Note that the detector 102 may use any method for detecting facial feature points. The detector 102 associates the detected facial feature points with the corresponding original images on a one-to-one basis. The detector 102 outputs the associated facial feature points and the original images to the crop unit 103 and the generation unit 105.

[0017] The crop unit 103 identifies a face in the original image associated with the facial feature points output by the detector 102. The crop unit 103 crops a partial image so that the entire identified face fits within the image. The crop unit 103 outputs the cropped partial image to the generation unit 105.

[0018] The second input unit 104 outputs to the generation unit 105 the shooting distance of each original image input to the detector 102 by the first input unit 101 and the shooting distance assumed for each new image to be newly generated from the original images. Note that the shooting distance assumed for each new image to be newly generated from the original images is longer than the shooting distance of the original images used for the generation.

[0019] As shown in FIG. 2, the generation unit 105 includes a calculation unit 1051, a first reproduction unit 1052, and a second reproduction unit 1053.

[0020] The calculation unit 1051 receives the shooting distance of each original image output from the second input unit 104, and the associated facial feature points and original images output from the detector 102. Then, based on the shooting distance of the original image received from the second input unit 104 and the facial feature points received from the detector 102, the calculation unit 1051 calculates the eye-to-eye distance for each partial image received from the crop unit 103 (i.e., the eye-to-eye distance for each face appearing in the original image). Specifically, the calculation unit 1051 identifies both eyes in each original image using the facial feature points. The calculation unit 1051 then determines how many pixels in the original image the identified number of pixels between the identified eyes is. Then, based on the shooting distance of the original image corresponding to the partial image received from the crop unit 103, the calculation unit 1051 calculates the distance corresponding to the identified number of pixels for the partial image received from the crop unit 103. The calculation unit 1051 outputs the calculated distance to the first reproduction unit 1052 as the eye distance.

[0021] The first reproduction unit 1052 generates an image by blurring the original image based on the eye distance output from the calculation unit 1051 and the assumed shooting distance for each new image to be newly generated from the original image output from the second input unit 104.

[0022] Specifically, the first reproduction unit 1052 reduces the resolution of the original image once to a resolution determined based on the eye distance output from the calculation unit 1051 and the shooting distance assumed for each new image to be newly generated from the original image output from the second input unit 104. Then, the first reproduction unit 1052 restores the image whose resolution has been reduced once to the resolution of the original image, thereby generating an image in which the original image is blurred.

[0023] FIG. 3 is a diagram illustrating processing performed by the first reproducing unit 1052 according to some embodiments of the present disclosure. Part (a) of FIG. 3 illustrates an original image of 300 × 300 pixels. Part (b) of FIG. 3 illustrates an image of 100 × 100 pixels obtained by reducing the resolution of the original image illustrated in part (a) of FIG. 3 to one-third. Part (c) of FIG. 3 illustrates an image obtained by restoring the image illustrated in part (b) of FIG. 3 to the resolution of the original image (i.e., returning it to an image of 300 × 300 pixels). For example, assume that the original image is the original image of 300 × 300 pixels illustrated in part (a) of FIG. 3. Furthermore, assume that the resolution calculated by the first reproducing unit 1052 based on the eye-to-eye distance calculated by the calculation unit 1051 is, for example, one-third of the resolution of the original image. In this case, the first reproducing unit 1052 reduces the resolution of the original image of 300 × 300 pixels shown in part (a) of Fig. 3 to one-third, thereby generating the image of 100 × 100 pixels shown in part (b) of Fig. 3. Then, the first reproducing unit 1052 restores the resolution of the generated 100 × 100 pixel image shown in part (b) of Fig. 3 to the resolution of the original image, thereby generating the image of 300 × 300 pixels shown in part (c) of Fig. 3. In this way, the first reproducing unit 1052 can generate an image in which the original image is blurred by once reducing the resolution of the original image to the resolution corresponding to the eye distance calculated by the calculation unit 1051 and then restoring it to the resolution of the original image.

[0024] Here, we will explain how the first reproduction unit 1052 determines the resolution based on the eye distance output from the calculation unit 1051 and the assumed shooting distance for each new image to be newly generated from the original image output from the second input unit 104.

[0025] When determining the resolution, the first reproduction unit 1052 uses a function represented by the curve in FIG. 4, which will be described below. First, a method for determining the function represented by the curve in FIG. 4 will be described. FIG. 4 is a diagram showing the correspondence relationship between eye distance and shooting distance. The correspondence relationship between eye distance and shooting distance is shown by the function represented by the curve in FIG. 4. Note that there is a one-to-one correspondence between the eye distance, the number of pixels corresponding to the eye distance, and the shooting distance. Therefore, for example, once the relationship between the shooting distance and the eye distance is determined, the corresponding number of pixels can be uniquely determined.

[0026] The correspondence relationship between the eye distance and the shooting distance shown in Figure 4 can be found, for example, by collecting data showing the relationship between the eye distance and the shooting distance and performing function fitting on the collected data. Furthermore, the correspondence relationship between the eye distance and the shooting distance shown in Figure 4 can be found, for example, based on a pinhole camera model. When found based on the pinhole camera model, the relationship shown in Equation (1) holds approximately.

[0027]

[0028] In equation (1), Δx is the coordinate shift in the image (i.e., the amount of change in eye-to-eye distance). Z is the shooting distance. ΔZ is the amount of change in shooting distance from Z. λ is a constant. Δx in equation (1) corresponds to the number of pixels. Therefore, the relationship between shooting distance and the number of pixels can be determined from equation (1). As described above, there is a one-to-one correspondence between eye-to-eye distance, the number of pixels corresponding to the eye-to-eye distance, and shooting distance. Therefore, once the relationship between shooting distance and the number of pixels is determined, the corresponding eye-to-eye distance can be uniquely determined. Therefore, by determining the relationship between shooting distance and the number of pixels from equation (1), the relationship between shooting distance and eye-to-eye distance can also be determined.

[0029] Next, a specific method by which the first reproduction unit 1052 determines the resolution will be described. Here, for example, assume that the shooting distance of the original image is 1 m, and a blurred image generated from the original image corresponds to a shooting distance of 3 m. In this case, based on the correspondence relationship between the eye-to-eye distance and the shooting distance shown in FIG. 4 , the first reproduction unit 1052 identifies point A corresponding to the original image and point B corresponding to the blurred image to be generated in FIG. 4 . As described above, there is a one-to-one correspondence between the eye-to-eye distance, the number of pixels corresponding to the eye-to-eye distance, and the shooting distance. Therefore, based on the relationship between the shooting distance and the eye-to-eye distance, the first reproduction unit 1052 identifies the number of pixels corresponding to point A as α pixels (100 pixels in the example shown in FIG. 4 ). Furthermore, based on the relationship between the shooting distance and the eye-to-eye distance, the first reproduction unit 1052 identifies the number of pixels corresponding to point B as β pixels (50 pixels in the example shown in FIG. 4 ). The first reproducing unit 1052 then obtains the conversion magnification by dividing β by α (in the example shown in FIG. 4, 50 is divided by 100 to obtain 1 / 2). The first reproducing unit 1052 uses this conversion magnification β / α as a magnification (1 / 3 in the example shown in FIG. 3) for reducing the resolution of the original image shown in part (a) of FIG. 3 described with reference to FIG. 3, thereby generating a blurred image as described with reference to FIG. 3. The first reproducing unit 1052 then outputs the generated blurred image to the second reproducing unit 1053.

[0030] The second reproduction unit 1053 adds noise to the blurred image generated by the first reproduction unit 1052. First, the noise added to the image will be described. FIG. 5 is a diagram illustrating an example of an original image according to some embodiments of the present disclosure. The original image illustrated in part (a) of FIG. 5 is an image captured at a relatively long shooting distance (i.e., a long-distance image). The original image illustrated in part (b) of FIG. 5 is an image captured at a relatively short shooting distance (i.e., a close-distance image) converted into a long-distance image. In other words, the original image illustrated in part (b) of FIG. 5 is an image to which blurring has been applied after processing by the first reproduction unit 1052. As can be seen from FIG. 5, the long-distance image has coarser noise than the image obtained by converting a close-distance image into a long-distance image. This shows that the coarseness of the noise depends on whether the original image is a long-distance image or a close-distance image (which ultimately depends on the eye-to-eye distance). Therefore, it can be seen that noise needs to be added when generating a new image from an original image. The second reproduction unit 1053 adds this noise. FIG. 6 illustrates an example of an image to which noise has been added according to some embodiments of the present disclosure. The image shown in (a) of FIG. 6 is an image to which noise has been added after conversion to a relatively low-resolution image. The image shown in (b) of FIG. 6 is an image to which noise has been added after conversion to a relatively high-resolution image. Note that the noise added to the low-resolution image and the noise added to the high-resolution image are similar. As can be seen from FIG. 6 , even when similar noise is added, the image after noise addition is significantly different between the low-resolution image and the high-resolution image. In other words, when adding noise to a blurred image generated by the first reproduction unit 1052, the second reproduction unit 1053 must convert the image to an image with an appropriate resolution before adding noise. Therefore, the appropriate resolution for adding noise to an image is determined in advance by performing experiments or simulations in which images with various resolutions are converted for each inter-eye distance and target shooting distance, and noise is added to the converted images.

[0031] The second reproduction unit 1053 receives the blurred image output by the first reproduction unit 1052. The second reproduction unit 1053 converts the received blurred image into an image with an appropriate resolution determined in advance according to the eye distance and the shooting distance to be converted. The second reproduction unit 1053 then adds noise to the converted image. The second reproduction unit 1053 outputs the image with the added noise to the resizing unit 106.

[0032] Examples of the type of noise added by the second reproduction unit 1053 include a uniform distribution within a fixed range, random numbers generated from a Gaussian distribution, etc. The intensity of the noise added by the second reproduction unit 1053 may be calculated from image information (average luminance, variance, etc.).

[0033] The resizing unit 106 receives the image output by the second reproduction unit 1053 (i.e., the image with noise added). The resizing unit 106 resizes the image with noise added received from the second reproduction unit 1053 to match the input size of the classifier 107. This resized image is a new image. The resizing unit 106 outputs the generated new image to the classifier 107. Note that a label suitable for classifying people based on their faces is assigned to the new image. Examples of labels include a label that permits a predetermined action and a label that prohibits a predetermined action. Examples of predetermined actions include passing through a gate installed in a facility, using financial facilities (e.g., withdrawing cash from an ATM (Automatic Teller Machine) or using a safe deposit box), etc.

[0034] The classifier 107 receives a new image to which a label suitable for classifying people based on their faces has been assigned, and determines weighting parameters (internal parameters of the classifier 107) by backpropagation according to the value of the evaluation function. Alternatively, the classifier 107 may receive a new image as a single large image, and determine (i.e., learn) the internal parameters of the classifier 107 by batch processing.

[0035] Furthermore, facial images of people captured by a camera installed in a facility or the like are input in real time to the trained classifier 107. Then, for example, the trained classifier 107 classifies the images into whether the person is permitted to perform a predetermined action or whether the person is prohibited from performing a predetermined action. The classifier 107 outputs the classification result to the control device 20.

[0036] 7 is a diagram illustrating an example of the configuration of the control device 20 according to some embodiments of the present disclosure. As illustrated in FIG. 7, the control device 20 includes a determination unit 201, a control unit 202, and a storage unit 203.

[0037] The determination unit 201 determines whether the classification result output by the trained classifier 107 indicates a person who is permitted to perform a predetermined action or a person who is prohibited from performing a predetermined action. The determination unit 201 outputs the determination result to the control unit 202.

[0038] The control unit 202 performs predetermined control when the determination unit 201 determines that the classification result by the trained classifier 107 indicates a person who is permitted to perform a predetermined action. Examples of the predetermined control include control to open a gate installed in a facility, and control to enable use of financial facilities (e.g., withdrawing cash from an ATM (Automatic Teller Machine) or using a safe deposit box). Specifically, for example, when the determination unit 201 determines that the classification result by the processing device 10 indicates a person who is permitted to perform a predetermined action (e.g., passing through a gate), the control unit 202 outputs a control signal to a device that drives a motor that opens and closes the gate, causing the gate to open. Also, specifically, for example, when the determination unit 201 determines that the classification result by the processing device 10 indicates a person who is permitted to perform a predetermined action (e.g., withdrawing cash from an ATM (Automatic Teller Machine)), the control unit 202 outputs a control signal to the ATM to permit withdrawal, including withdrawal of cash. Specifically, for example, when the determination unit 201 determines that the classification result by the processing device 10 indicates a person who is permitted to perform a predetermined action (for example, using a safe deposit box), the control unit 202 outputs a control signal to a device that manages the safe deposit box to permit use of the safe deposit box. The processing device 10 is a device used together with a biometric authentication system, and detects spoofing of the biometric authentication system (for example, pretending to be another person by holding a photo or the like up to a camera to deceive the biometric authentication system).

[0039] Note that the processing performed by the processing system 1 in each embodiment of the present disclosure is not limited to the above-described processing. For example, the processing system 1 may perform the processing described below.

[0040] (Processing Performed by Processing System) Fig. 8 is a diagram showing a first example of a processing flow of the processing system 1 according to some embodiments of the present disclosure. Fig. 9 is a diagram showing a second example of a processing flow of the processing system 1 according to some embodiments of the present disclosure. The processing flow shown in Fig. 8 shows processing in which new images for training the classifier 107 are generated from original images, and the classifier 107 trains using the newly generated new images. The processing flow shown in Fig. 9 shows processing in which the control device 20 performs predetermined control based on the classification results output by the trained classifier 107.

[0041] First, the processing performed by the processing system 1 will be described with reference to Fig. 8. The first input unit 101 inputs a plurality of original images to the detector 102. The detector 102 detects facial feature points in each of the original images (step S1). The detector 102 outputs the associated facial feature points and the original image to the crop unit 103 and the generation unit 105.

[0042] The crop unit 103 identifies a face in the original image associated with the facial feature points output by the detector 102 (step S2). The crop unit 103 crops a partial image so that the entire identified face fits in the partial image (step S3). The crop unit 103 outputs the cropped partial image to the generation unit 105.

[0043] The second input unit 104 outputs to the generation unit 105 the shooting distance of each original image input to the detector 102 by the first input unit 101 and the shooting distance assumed for each new image to be newly generated from the original images. Note that the shooting distance assumed for each new image to be newly generated from the original images is longer than the shooting distance of the original images used for the generation.

[0044] The calculation unit 1051 receives the shooting distance of each original image output from the second input unit 104, and the associated facial feature points and original images output from the detector 102. Then, the calculation unit 1051 calculates the eye-to-eye distance for each partial image received from the crop unit 103 (i.e., the eye-to-eye distance for each face captured in the original image) based on the shooting distance of the original image received from the second input unit 104 and the facial feature points received from the detector 102 (step S4). The calculation unit 1051 outputs the calculated eye-to-eye distance to the first reproduction unit 1052.

[0045] The first reproduction unit 1052 generates an image by blurring the original image based on the eye distance output from the calculation unit 1051 and the shooting distance assumed for each new image to be newly generated from the original image output from the second input unit 104 (step S5). The first reproduction unit 1052 outputs the generated blurred image to the second reproduction unit 1053.

[0046] The second reproduction unit 1053 receives the blurred image output by the first reproduction unit 1052. The second reproduction unit 1053 converts the received blurred image into an image with a predetermined appropriate resolution according to the eye distance and the target shooting distance (step S6). The second reproduction unit 1053 then adds noise to the converted image (step S7). The second reproduction unit 1053 outputs the image with added noise to the resizing unit 106.

[0047] Examples of the type of noise added by the second reproduction unit 1053 include a uniform distribution within a fixed range, random numbers generated from a Gaussian distribution, etc. The intensity of the noise added by the second reproduction unit 1053 may be calculated from image information (average luminance, variance, etc.).

[0048] The resizing unit 106 receives the image output by the second reproduction unit 1053 (i.e., the image with noise added). The resizing unit 106 resizes the image with noise added received from the second reproduction unit 1053 to match the input size of the classifier 107 (step S8). A new image is generated by the processing of step S8. The resizing unit 106 outputs the generated new image to the classifier 107.

[0049] The classifier 107 receives as input a new image to which a label suitable for classifying people based on their faces has been assigned, and determines weighting parameters (internal parameters of the classifier 107) by backpropagation according to the value of the evaluation function (step S9). The classifier 107 that has completed the processing of step S9 is the trained classifier 107. Note that the classifier 107 may receive a new image as a single large image and determine the internal parameters of the classifier 107 by batch processing.

[0050] Next, the processing performed by the processing system 1 will be described with reference to Fig. 9. For example, facial images of people captured by a camera installed in a facility or the like are input in real time to the trained classifier 107. Then, for example, the trained classifier 107 classifies the people into those who are permitted to perform a predetermined action and those who are prohibited from performing a predetermined action (step S11). The classifier 107 outputs the classification result to the control device 20.

[0051] The determination unit 201 determines whether the classification result output by the trained classifier 107 indicates a person who is permitted to perform a predetermined action or a person who is prohibited from performing a predetermined action (step S12). The determination unit 201 outputs the determination result to the control unit 202.

[0052] When the determination unit 201 determines that the classification result by the trained classifier 107 indicates a person who is permitted to perform a predetermined action, the control unit 202 performs a predetermined control (step S13).

[0053] (Advantages) The processing system 1 according to an embodiment of the present disclosure has been described above. In the processing device 10 of the processing system 1, the generation unit 105 (an example of a processing means) generates a new image for training a classifier that classifies images, based on the eye-to-eye distance of a face included in an original image and an assumed shooting distance for a new image generated from the original image. This processing device 10 can generate new images for training a classifier from existing images.

[0054] In the processing system 1 according to another embodiment of the present disclosure, the assumed shooting distance for each new image to be newly generated from the original image output by the second input unit 104 to the generation unit 105 may be determined randomly. FIG. 10 is a diagram illustrating an example of a configuration of a processing device 10 according to another embodiment of the present disclosure. As shown in FIG. 10 , the processing device 10 of the processing system 1 includes a first input unit 101, a detector 102, a crop unit 103, a second input unit 104, a generation unit 105, and a classifier 107. Also, as shown in FIG. 10 , the processing device 10 includes a switching unit 108. The processing device 10 differs from the processing device 10 according to an embodiment of the present disclosure in the crop unit 103, the second input unit 104, the resizing unit 106, and the switching unit 108.

[0055] The Crop unit 103 identifies a face in the original image associated with the facial feature points output by the detector 102. The Crop unit 103 cuts out a partial image so that the entire identified face fits in the image. The Crop unit 103 outputs the cut-out partial image to the generation unit 105 and the switching unit 108.

[0056] The second input unit 104 randomly determines an assumed shooting distance Z for each new image to be generated from the original image, between Z_low < Z < Z_high (for example, Z_low is 1 m and Z_high is 3 m). Then, the second input unit 104 outputs the determined shooting distance L to the generation unit 105.

[0057] The switching unit 108 selects the cropped partial image output by the crop unit 103 and the new image output by the generation unit 105, and outputs the selected image to the resizing unit 106. For example, a probability P at which the generation unit 105 generates a new image is determined in advance, and when the generation unit 105 is in a state where it will generate a new image with the probability P, the switching unit 108 selects the new image output by the generation unit 105 and outputs the selected image to the resizing unit 106. When the generation unit 105 is in a state other than the above with the probability (1-P), the switching unit 108 selects the cropped partial image output by the crop unit 103 and outputs the selected image to the resizing unit 106.

[0058] The resizing unit 106 receives the image output by the switching unit 108. The resizing unit 106 resizes the image received from the switching unit 108 to match the input size of the classifier 107. This resized image is a new image. The resizing unit 106 outputs the generated new image to the classifier 107.

[0059] Such a processing system 1 can generate new images corresponding to various shooting distances.

[0060] In addition, in the processing system 1 according to another embodiment of the present disclosure, the assumed shooting distance for each new image to be newly generated from the original image output by the second input unit 104 to the generation unit 105 may be determined so that the number of new images corresponding to each shooting distance is uniform.

[0061] 2 or 10 , the second input unit 104 determines the shooting distance Z assumed for each new image to be newly generated from the original image so that the number of new images corresponding to each shooting distance is uniform within Z_low < Z < Z_high (for example, Z_low is 1 m and Z_high is 3 m). Then, the second input unit 104 outputs the determined shooting distance Z to the generation unit 105.

[0062] Such a processing system 1 can generate new images corresponding to various shooting distances.

[0063] A processing device 10 according to another embodiment of the present disclosure will be described. FIG. 11 is a diagram illustrating an example of the configuration of the processing device 10 according to another embodiment of the present disclosure. As shown in FIG. 11 , the processing device 10 according to another embodiment of the present disclosure includes processing means 300. The processing means 300 generates a new image for training a classifier that classifies images based on the eye-to-eye distance of a face included in an original image and an assumed shooting distance for a new image generated from the original image. The processing means 300 can be realized, for example, using the functions of the generation unit 105 exemplified in FIGS. 2 and 10 .

[0064] Next, a description will be given of processing by the processing device 10. Fig. 12 is a diagram showing an example of a processing flow of the processing device 10 according to another embodiment of the present disclosure. Here, the processing by the processing device 10 will be described with reference to Fig. 12.

[0065] The processing means 300 generates a new image for training a classifier that classifies images based on the eye distance of a face included in the original image and an assumed shooting distance for a new image generated from the original image (step S101). In this way, the processing device 10 can generate a new image for training a classifier from an existing image.

[0066] The processing device 10 according to each embodiment of the present disclosure may be applied to a face authentication device that uses a face image to detect impersonation of a person.

[0067] The order of the processes in the embodiments of the present disclosure may be changed as long as the processes are performed appropriately.

[0068] Although the embodiments of the present disclosure have been described, the processing system 1, the processing device 10, the control device 20, and other control devices may have a computer device inside. The above-described processing steps are stored in the form of a program on a computer-readable recording medium, and the computer reads and executes this program to perform the above processing. Specific examples of computers are shown below.

[0069] 13 is a schematic block diagram showing the configuration of a computer according to at least one embodiment. As shown in FIG. 13 , the computer 5 includes a CPU (Central Processing Unit) 6, a main memory 7, a storage 8, and an interface 9. For example, the processing system 1, the processing device 10, the control device 20, and other control devices described above are each implemented in the computer 5. The operations of each of the processing units described above are stored in the storage 8 in the form of a program. The CPU 6 reads the program from the storage 8, loads it into the main memory 7, and executes the above-described processing in accordance with the program. The CPU 6 also allocates storage areas in the main memory 7 corresponding to each of the storage units described above in accordance with the program.

[0070] Examples of storage 8 include a hard disk drive (HDD), a solid state drive (SSD), a magnetic disk, a magneto-optical disk, a compact disc read-only memory (CD-ROM), a digital versatile disc read-only memory (DVD-ROM), and a semiconductor memory. Storage 8 may be an internal medium directly connected to the bus of computer 5, or an external medium connected to computer 5 via interface 9 or a communication line. Furthermore, if the program is distributed to computer 5 via a communication line, computer 5 that receives the program may load the program into main memory 7 and execute the above-described processing. In at least one embodiment, storage 8 is a non-transitory tangible storage medium.

[0071] The program may also implement some of the functions described above. Furthermore, the program may be a file that can implement the functions described above in combination with a program already stored in the computer device, a so-called differential file (differential program).

[0072] Although several embodiments of the present disclosure have been described, these embodiments are merely examples and do not limit the scope of the disclosure. Various additions, omissions, substitutions, and modifications may be made to these embodiments without departing from the spirit of the disclosure.

[0073] Note that part or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0074] (Supplementary Note 1) A processing device comprising: a processing means for generating a new image for training a classifier that classifies images based on an inter-eye distance of a face included in an original image and an assumed shooting distance for the new image to be generated from the original image.

[0075] (Supplementary Note 2) The processing device according to Supplementary Note 1, wherein the new image is an image having a lower resolution than the original image.

[0076] (Supplementary Note 3) The processing device according to Supplementary Note 1 or Supplementary Note 2, wherein the processing means comprises: a first processing means that adds noise to an image of a predetermined resolution in the middle of generating the new image in order to generate the new image.

[0077] (Supplementary Note 4) The processing device described in any one of Supplementary Note 1 to Supplementary Note 3, wherein the processing means is a second processing means that converts an image in the middle of generating the new image to a relatively lower resolution in order to generate the new image, and returns the resolution of the converted image to the resolution of the image before conversion.

[0078] (Supplementary Note 5) The processing device according to any one of Supplementary Note 1 to Supplementary Note 4, further comprising: a third processing means for classifying an image, wherein an internal parameter is determined based on the original image and the new image.

[0079] (Supplementary Note 6) A processing system comprising: the processing device according to any one of Supplementary Note 1 to Supplementary Note 5; and a control device that performs predetermined control based on a classification result of the image by the processing device.

[0080] (Supplementary Note 7) A processing method for generating a new image for training a classifier that classifies images, based on the distance between the eyes of a face included in an original image and an assumed shooting distance for the new image generated from the original image.

[0081] (Supplementary Note 8) A recording medium storing a program for causing a computer to execute the steps of: generating a new image for training a classifier that classifies images based on the distance between the eyes of a face included in an original image and an assumed shooting distance for the new image generated from the original image.

[0082] According to aspects of the present disclosure, new images can be generated from existing images to train a classifier.

[0083] REFERENCE SIGNS LIST 1 Processing system 5 Computer 6 CPU 7 Main memory 8 Storage 9 Interface 10 Processing device 20 Control device 101 First input unit 102 Detector 103 Crop unit 104 Second input unit 105 Generation unit 106 Resize unit 107 Classifier 108 Switching unit 201 Determination unit 202 Control unit 300 Processing means 1051 Calculation unit 1052 First reproduction unit 1053 Second reproduction unit

Claims

1. A processing device comprising: a processing means for generating a new image for training a classifier that classifies images based on the eye distance of a face contained in an original image and an assumed shooting distance for the new image to be generated from the original image.

2. The processing device according to claim 1, wherein the new image is an image having a lower resolution than the original image.

3. A processing device according to claim 1 or claim 2, wherein the processing means comprises: first processing means for adding noise to an image of a predetermined resolution in the process of generating the new image in order to generate the new image.

4. A processing device as claimed in any one of claims 1 to 3, wherein the processing means comprises: second processing means which converts an image in the middle of generating the new image to a relatively lower resolution in order to generate the new image, and returns the resolution of the converted image to the resolution of the image before conversion.

5. The processing device according to any one of claims 1 to 4, further comprising: a third processing means for classifying an image, the third processing means determining internal parameters based on the original image and the new image.

6. A processing system comprising: a processing device according to any one of claims 1 to 5; and a control device that performs predetermined control based on the image classification result by said processing device.

7. A processing method for generating a new image for training a classifier that classifies images based on the eye distance of a face contained in an original image and an assumed shooting distance for the new image generated from the original image.

8. A recording medium storing a program for causing a computer to execute the following: generating a new image for training a classifier that classifies images based on the distance between the eyes of a face included in an original image and an assumed shooting distance for the new image generated from the original image.

Images