Information processing device, information processing method, and computer program
Patent Information
- Application Number
- JP2025508011
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-07-22
- Publication Date
- 2025-10-01
- Estimated Expiration
- 2043-03-22
Abstract
Description
Information processing device, information processing method, and recording medium
[0001] The present disclosure relates to the technical fields of an information processing device, an information processing method, and a recording medium.
[0002] Patent Document 1 describes a technology that stores information about a classifier that outputs the part-likelihood of a specific part of a detection target, determines the part-likelihood of an image region included in an input image based on the classifier information, determines the position of the specific part in the input image based on the determined part-likelihood, stores information about a reference position of the specific part, determines difference information between the reference position of the specific part and the determined position of the specific part based on the information about the reference position, and determines a confidence level indicating the possibility that the input image is an image of the detection target based on the difference information.
[0005] Patent Literature 2 describes a technique for generating a face direction conversion image by converting the face direction of a face shown in an input image to a predetermined direction according to each of a plurality of assumed face directions, inverting the face shown in the face direction conversion image for each assumed face direction to generate an inverted face image, converting the face direction of the inverted face image to the assumed face direction or converting the face direction of the input image to a predetermined direction according to the assumed face direction, calculating an evaluation value representing a degree of difference between the face shown in the inverted face image and the face shown in the input image for each assumed face direction based on the conversion result, and identifying one of the assumed face directions as the face direction shown in the input image based on the evaluation value.
[0006] Patent Literature 3 describes a technique for acquiring data related to a detection target, extracting common features common to a plurality of candidates for an attribute of the detection target from the data, detecting feature information for each of the plurality of candidates based on the common features, determining the attribute of the detection target based on the data, and outputting feature information corresponding to the determined attribute. Non-Patent Document 1 describes a technology that uses a large number of annotated face-less images to train a network that reconstructs faces using an hourglass-type network autoencoder, then fixes the weights of the autoencoder and adds a branching network along the way to train feature point detection using a small number of annotated face images.
[0003] International Publication No. 2013 / 122009 Japanese Patent Application Laid-Open No. 2018-097573 International Publication No. 2022 / 003982
[0004] Bjorn Browatzki and Christian Wallraven. "3fabrec: Fast few-shot face alignment by reconstruction." Proceedings of the IEEE / CVF Conference on Computer Vision and Pattern Recognition. 2020.
[0005] An object of this disclosure is to provide an information processing device, an information processing method, and a recording medium that aim to improve upon the techniques described in prior art documents.
[0006] One aspect of the information processing device includes an acquisition means for acquiring a correct-answer image having correct correct-answer feature information and a correct-answer non-correct image having no correct feature information, and a learning means for causing the detection mechanism to learn a detection method for detecting feature information from an input image using the correct-answer feature information detected by the detection mechanism from the correct-answer image and correct-answer detection information based on the correct feature information, as well as correct-answer non-feature information detected by the detection mechanism from the correct-answer non-image and correct-answer non-detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the correct-answer non-image.
[0007] One aspect of the information processing method involves acquiring a correct-answer image having correct correct-answer feature information and a correct-answer non-correct image having no correct feature information, and having the detection mechanism learn a detection method for detecting feature information from an input image using the correct-answer feature information detected by the detection mechanism from the correct-answer image and correct-answer detection information based on the correct feature information, as well as correct-answer non-feature information detected by the detection mechanism from the correct-answer non-image and correct-answer non-detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the correct-answer non-image.
[0008] One aspect of the recording medium has recorded thereon a computer program for causing a computer to execute an information processing method in which a correct-answer image having correct correct-answer feature information and a correct-answer no-answer image not having correct feature information are acquired, and the detection mechanism learns a detection method for detecting feature information from an input image using the correct-answer feature information detected by the detection mechanism from the correct-answer image and correct-answer detection information based on the correct answer feature information, as well as the correct-answer no-answer feature information detected by the detection mechanism from the correct-answer no-answer image and correct-answer no-answer detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the correct-answer no-answer image.
[0009] FIG. 1 is a block diagram showing the configuration of an information processing device in a first embodiment. FIG. 2 is a block diagram showing the configuration of a learning device in a second embodiment. FIG. 3 is a flowchart showing the flow of learning operations of the learning device in the second embodiment. FIG. 4 is a conceptual diagram showing the flow of learning operations of the learning device in the second embodiment. FIG. 5 is a block diagram showing the configuration of a detection device in a third embodiment. FIG. 6 is a flowchart showing the flow of detection operations of the detection device in the third embodiment. FIG. 7 is a block diagram showing the configuration of an authentication device in a fourth embodiment. FIG. 8 is a flowchart showing the flow of authentication operations of the authentication device in the fourth embodiment.
[0010] Hereinafter, embodiments of an information processing device, an information processing method, and a recording medium will be described with reference to the drawings. [1: First Embodiment]
[0011] A first embodiment of an information processing device, an information processing method, and a recording medium will be described below. Hereinafter, the first embodiment of the information processing device, the information processing method, and the recording medium will be described using an information processing device 1 to which the first embodiment of the information processing device, the information processing method, and the recording medium is applied. [1-1: Configuration of Information Processing Device 1]
[0012] FIG. 1 is a block diagram showing the configuration of an information processing device 1 according to the first embodiment. As shown in FIG. 1, the information processing device 1 includes an acquisition unit 11 and a learning unit 12. The acquisition unit 11 acquires correct-answer images having correct correct-answer feature information and correct-answer non-images having no correct feature information. The learning unit 12 causes the detection mechanism to learn a detection method for detecting feature information from an input image, using the correct-answer feature information detected by the detection mechanism from the correct-answer images and correct-answer detection information based on the correct answer feature information, as well as the correct-answer non-feature information detected by the detection mechanism from the correct-answer non-images and correct-answer non-detection information based on extended feature information detected by the detection mechanism from extended images obtained by extending the correct-answer non-images. [1-2: Technical Effects of the Information Processing Device 1]
[0013] The information processing device 1 in the first embodiment uses correct images having correct feature information and correct non-correct images having no correct feature information to train the detection mechanism to detect a detection method for detecting feature information from an input image. By using correct non-correct images having no correct feature information, which are relatively easy to prepare, in addition to correct images having correct feature information, the information processing device 1 can relatively easily train the detection mechanism to a detection method that can accurately detect feature information. [2: Second embodiment]
[0014] Next, a second embodiment of the information processing device, the information processing method, and the recording medium will be described. Hereinafter, the second embodiment of the information processing device, the information processing method, and the recording medium will be described using a learning device 2 to which the second embodiment of the information processing device, the information processing method, and the recording medium is applied. [2-1: Configuration of the learning device 2]
[0015] 2 is a block diagram showing the configuration of a learning device 2 in the second embodiment. As shown in FIG. 2, the learning device 2 includes a calculation device 21 and a storage device 22. The learning device 2 may further include a communication device 23, an input device 24, and an output device 25. However, the learning device 2 does not have to include at least one of the communication device 23, the input device 24, and the output device 25. The calculation device 21, the storage device 22, the communication device 23, the input device 24, and the output device 25 may be connected via a data bus 26.
[0016] The arithmetic device 21 includes, for example, at least one of a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and an FPGA (Field Programmable Gate Array). The arithmetic device 21 loads a computer program. For example, the arithmetic device 21 may load a computer program stored in the storage device 22. For example, the arithmetic device 21 may load a computer program stored in a computer-readable, non-transitory storage medium using a storage medium reading device (e.g., the input device 24 described later) not shown in the drawing provided in the learning device 2. The arithmetic device 21 may acquire (i.e., download or load) the computer program from a device (not shown) located outside the learning device 2 via the communication device 23 (or another communication device). The arithmetic device 21 executes the loaded computer program. As a result, logical functional blocks for executing the operations to be performed by the learning device 2 are realized within the arithmetic device 21. In other words, the arithmetic device 21 can function as a controller for realizing logical functional blocks for executing the operations (in other words, processing) to be performed by the learning device 2.
[0017] FIG. 2 shows an example of logical functional blocks implemented within the arithmetic device 21 to perform information processing operations. As shown in FIG. 3 , the arithmetic device 21 implements an acquisition unit 211, which is a specific example of an "acquisition means" described in the appendix below, and a learning unit 212, which is a specific example of a "learning means" described in the appendix below. The learning unit 212 may include a loss function calculation unit 2121 and a parameter update unit 2122. Furthermore, the arithmetic device 21 implements a detection unit 213, a conversion unit 214, and an inverse conversion unit 215. However, at least one of the detection unit 213, the conversion unit 214, and the inverse conversion unit 215 does not have to be implemented within the arithmetic device 21. Details of the operations of the acquisition unit 211, the learning unit 212, the detection unit 213, the conversion unit 214, and the inverse conversion unit 215 will be described later with reference to FIGS. 3 and 4 .
[0018] The storage device 22 can store desired data. For example, the storage device 22 may temporarily store a computer program executed by the arithmetic device 21. The storage device 22 may temporarily store data that the arithmetic device 21 temporarily uses when the arithmetic device 21 is executing a computer program. The storage device 22 may store data that the learning device 2 stores long-term. The storage device 22 may include at least one of a RAM (Random Access Memory), a ROM (Read Only Memory), a hard disk device, a magneto-optical disk device, an SSD (Solid State Drive), and a disk array device. In other words, the storage device 22 may include a non-transitory recording medium. A learning data storage unit 221 and a parameter storage unit 222 may be implemented within the storage device 22.
[0019] The learning data storage unit 221 may store correct images and non-correct images as learning data. Correct images have correct feature information. Correct images may be annotated images that are annotated with correct feature information. Non-correct images do not have correct feature information. Non-correct images may be non-annotated images that are not annotated with correct feature information.
[0020] For example, 1,000 correct images and 100,000 non-correct images may be stored as training data in the training data storage unit 221. The numbers 1,000 and 100,000 are merely examples, but the training data storage unit 221 may store an order of magnitude more non-correct images than correct images.
[0021] The communication device 23 can communicate with devices external to the learning device 2 via a communication network (not shown). The communication device 23 may be a communication interface based on standards such as Ethernet (registered trademark), Wi-Fi (registered trademark), Bluetooth (registered trademark), or USB (Universal Serial Bus).
[0022] Input device 24 is a device that accepts information input to learning device 2 from outside learning device 2. For example, input device 24 may include an operating device (e.g., at least one of a keyboard, a mouse, and a touch panel) that can be operated by an operator of learning device 2. For example, input device 24 may include a reading device that can read information recorded as data on a recording medium that can be attached externally to learning device 2.
[0023] The output device 25 is a device that outputs information to the outside of the learning device 2. For example, the output device 25 may output information as an image. That is, the output device 25 may include a display device (a so-called display) that can display an image showing the information to be output. For example, the output device 25 may output information as sound. That is, the output device 25 may include an audio device (a so-called speaker) that can output sound. For example, the output device 25 may output information on paper. That is, the output device 25 may include a printing device (a so-called printer) that can print desired information on paper. [2-2: Learning Operations Performed by the Learning Device 2]
[0024] The flow of the learning operation performed by the learning device 2 will be described with reference to Figures 3 and 4. Figure 3 is a flowchart showing the flow of the learning operation performed by the learning device 2. Figure 4 is a conceptual diagram of the learning operation performed by the learning device 2. [Detection mechanism DM]
[0025] The detection mechanism DM detects feature information from an input image. The detection mechanism DM in the second embodiment is a model trained to output facial feature points when a face image is input. The detection mechanism DM detects facial organ points such as the eyes, nose, and mouth. The detection mechanism DM may be, for example, a model that outputs the positions of both eyes, the tip of the nose, and both corners of the mouth as feature points, as shown in FIG. 4 .
[0026] Parameters necessary for the detection mechanism DM to function are stored in the parameter storage unit 222. For example, the parameter storage unit 222 stores parameters necessary for the detection mechanism DM to configure a neural network.
[0027] The detection mechanism DM includes a convolutional neural network and a fully connected layer. The convolutional neural network and the fully connected layer are connected in series. In the following description, we will not distinguish between the convolutional neural network and the fully connected layer.
[0028] The detection mechanism DM used in each of the network 1 illustrated in FIG. 4(a) and the network 2 illustrated in FIG. 4(b) has the same structure and performs learning using common weight parameters. That is, the detection mechanism DM includes a neural network configured with common weight parameters. Each of the network 1 and the network 2 uses the detection mechanism DM with common weight parameters. [Network 1: Operation using annotated image Ia]
[0029] 3, the learning unit 212 performs learning of the detection mechanism DM using the annotation-containing image Ia as the correct image (step S20). Fig. 4A is a conceptual diagram of the learning operation of the detection mechanism DM using the annotation-containing image Ia.
[0030] 4A, the acquisition unit 211 acquires an annotation-included image Ia. The annotation-included image Ia may be annotated with the correct coordinate values of the correct feature point FP1. The acquisition unit 211 may acquire the annotation-included image Ia from the training data storage unit 221.
[0031] The detection unit 213 detects annotation feature points FP2 from the annotation-included image Ia. The detection unit 213 inputs the annotation-included image Ia to the detection mechanism DM, and outputs the annotation feature points FP2 output by the detection mechanism DM as the detection result. When the feature points are defined as shown in FIG. 4, the detection mechanism DM may output ten numerical values indicating the x-coordinates and y-coordinates of the five feature points, respectively.
[0032] The learning unit 212 causes the detection mechanism DM to learn a feature point detection method using supervised detection information based on the supervised feature point FP1 and the annotated feature point FP2 detected by the detection mechanism DM from the annotated image Ia. The loss function calculation unit 2121-a calculates the loss La based on a loss function in which the loss increases the more dissimilar the supervised feature point FP1 and the annotated feature point FP2 are. This loss La is called the supervised feature point error loss La. For example, the loss function calculation unit 2121-a may calculate the supervised feature point error loss La based on the 10 coordinate values output when the annotated image Ia is input to the detection mechanism DM and the coordinate value of the supervised feature point FP1. The supervised feature point error loss La is a loss in regression learning, and is a loss that decreases as the information annotated on the input image and the detection result from the input image match more closely. [Network 2: Operation using an unannotated image Ib]
[0033] 3, the learning unit 212 performs learning of the detection mechanism DM using the annotation-free image Ib as the correct image (step S21). Fig. 4B is a conceptual diagram of the learning operation of the detection mechanism DM using the annotation-free image Ib.
[0034] As shown in FIG. 4B, the acquisition unit 211 acquires an annotation-free image Ib. Hereinafter, the annotation-free image Ib will be referred to as the original image Ib. The detection unit 213 detects original feature points FP3 from the original image Ib. The detection unit 213 inputs the original image Ib to a detection mechanism DM, and outputs the original feature points FP3 output by the detection mechanism DM as the detection result.
[0035] The conversion unit 214 applies a conversion to the original image Ib to generate an expanded image Ic by expanding the original image Ib. For example, as illustrated in Fig. 4B, the conversion unit 214 may generate the expanded image Ic by flipping the original image Ib horizontally. Examples of the expansion of the input image by the conversion unit 214 include inversion of the input image, such as horizontal flipping, as illustrated in Fig. 4B, as well as translation of the input image and rotation of the input image.
[0036] The detection unit 213 detects an extended feature point FP4 from the extended image Ic. The detection unit 213 inputs the extended image Ic to the detection mechanism DM, and outputs the extended feature point FP4 output by the detection mechanism DM as the detection result.
[0037] The inverse transformation unit 215 performs on the extended feature point FP4 the inverse transformation of the transformation performed on the original image Ib by the transformation unit 214. For example, as illustrated in FIG. 4B , the inverse transformation unit 215 may generate an inversely transformed feature point FP5 by inverting the extended feature point FP4.
[0038] The learning unit 212 uses correct non-detection information based on the original feature points FP3 and the inverse-transformed feature points FP5 to train the detection mechanism DM on a feature point detection method. The loss function calculation unit 2121-b calculates the loss Lb based on a loss function in which the loss increases the more dissimilar the original feature points FP3 and the inverse-transformed feature points FP5 are. This loss Lb is called the unsupervised invariance loss Lb. For example, the loss function calculation unit 2121-b may calculate the unsupervised invariance loss Lb based on 10 coordinate values output when the original image Ib is input to the detection mechanism DM and coordinate values obtained by inversely transforming the 10 coordinate values output when the augmented image Ic is input to the detection mechanism DM. The unsupervised invariance loss Lb is a loss that decreases as the detection results without data augmentation and the detection results with data augmentation match. The agreement of results with and without data augmentation can be said to be a necessary condition for both systems to correctly detect feature points.
[0039] 4B, it is possible to learn the same number of times for the face facing right and the face facing left, which prevents a bias in the detection ability, such as being good at detecting the face facing right or the face facing left, but not the other way around.
[0040] Note that steps S20 and S21 may be performed in reverse order or simultaneously. [Updating Parameters of Detection Mechanism DM]
[0041] The parameter update unit 2122 updates the weight parameters of the detection mechanism so that the supervised feature point error loss La and the unsupervised invariance loss Lb are minimized. The parameter update unit 2122 may update the weight parameters of the detection mechanism so that L expressed in the following equation 1 is minimized: L=La+λLb (Equation 1)
[0042] The algorithm for determining the above parameters so as to minimize the supervised feature point error loss La and the unsupervised invariance loss Lb may be any learning algorithm used in machine learning, such as gradient descent or backpropagation.
[0043] The parameter update unit 2122 may perform a parameter update operation each time a detection operation is performed on a predetermined number of face images. The predetermined number of face images may include correct images and non-correct images at the same ratio as the ratio of all face images stored in the learning data storage unit 221. For example, if the ratio of correct images to non-correct images is 1:10, the predetermined number of face images may include non-correct images in a number 10 times the number of correct images. For example, if the ratio of correct images to non-correct images is 1:10, λ in the above formula 1 may be set to "0.1." In other words, the loss L may be calculated so that the supervised feature point error loss La and the unsupervised invariance loss Lb contribute equally.
[0044] Alternatively, the learning unit 212 may perform learning using the annotation-free image Ib and then perform learning using the annotation-containing image Ia. Conversely, the learning unit 212 may perform learning using the annotation-containing image Ia and then perform learning using the annotation-free image Ib.
[0045] It is determined whether learning has ended (step S22). For example, the learning unit 212 may determine that learning has ended when learning has been performed using all face images stored in the learning data storage unit 221. In another example, the learning unit 212 may determine that learning has ended when L expressed in the above formula 1 is equal to or less than a predetermined threshold. Alternatively, the learning unit 212 may determine that learning has ended when both the losses calculated in steps S20 and S21 are equal to or less than a predetermined threshold. In yet another example, the learning unit 212 may determine that learning has ended when steps S20 and S21 have been repeated a predetermined number of times.
[0046] If the learning unit 212 determines that the learning has been completed (step S22; Yes), the learning unit 212 determines the parameters of the detection mechanism DM and determines the detection mechanism DM (step S23). The learning unit 212 may transmit the determined parameters of the detection mechanism DM to a device that performs a detection operation using the detection mechanism DM. On the other hand, if the learning unit 212 determines that the learning has not been completed (step S22; No), the processing returns to step S20. [2-3: Modification]
[0047] In the second embodiment, the feature information of an image is explained as an example of feature points of the image. However, the feature information may be information of features other than feature points. For example, the feature information of an image may be information indicating a feature region within the image. The information indicating the feature region may be information used for semantic segmentation or the like. The learning operation in the second embodiment may be widely applied to tasks that use images as input.
[0048] Furthermore, in the second embodiment, the image is a face image, but the image may be an image other than a face. For example, the image may be a body image of a living creature. The living creature may be a mammal, including a human, a bird, or the like. Alternatively, the image may be an image including numbers, such as a license plate number of a vehicle.
[0049] For example, in image recognition, technology for detecting feature points with high accuracy is important. Feature points detected from a face can be used, for example, for face authentication, facial expression recognition, etc. Feature points detected from a body can be used, for example, for estimating a human body pose, analyzing a person's behavior, and matching a human body. Feature points detected from a body may be, for example, joint points. [2-4: Technical Effects of Learning Device 2]
[0050] The learning device 2 in the second embodiment learns a detection method such that correct feature information and detected feature information are similar, and also learns a detection method such that feature information detected from an image and feature information detected from an extended image are similar, so it is possible to learn a detection method that accurately detects feature information.The learning device 2 learns a detection method such that feature information detected from an extended image obtained by converting and extending an original image is inversely converted and that feature information detected from the original image is similar, so it is possible to learn a detection method that accurately detects feature information.
[0051] In order to accurately detect feature points from images, it is common to prepare a large number of annotated images, which have been annotated with the feature points of the image, for training. Preparing a large number of annotated images is a time-consuming and labor-intensive task. If feature points can be accurately detected from images by training using a small number of annotated images, it would be possible to reduce labor costs and time.
[0052] The learning device 2 uses images without annotations, which are relatively easy to prepare, in addition to images with annotations, allowing the detection mechanism to relatively easily learn a detection method that can accurately detect feature information. Because the learning device 2 uses a relatively large number of images without annotations, which are relatively easy to prepare, it is possible to reduce the personnel costs and time required to prepare learning data. The learning device 2 can use a small number of images with annotations to allow the detection mechanism to learn a detection method that can accurately detect feature points.
[0053] Non-Patent Document 1 above describes a comparative example that uses a small number of annotated face images and a large number of annotated face images. The comparative example described in Non-Patent Document 1 uses an hourglass network that requires an encoder and a decoder. In other words, there are limitations to the network. The processing of the hourglass network requires a large amount of calculation, is computationally inefficient, and has slow learning and inference speeds. Furthermore, the comparative example requires relatively heavy processing because feature points are found from the peaks of the heat map of each feature point output by the hourglass network.
[0054] In contrast, the detection mechanism DM in this embodiment directly outputs feature information without using a heat map. Since the detection mechanism DM in this embodiment does not use an hourglass network, the amount of calculation can be reduced, resulting in good calculation efficiency. Furthermore, since the detection mechanism DM in this embodiment directly outputs feature information, the processing is lighter than the processing in the comparative example. [3: Third Embodiment]
[0055] Next, a third embodiment of the information processing device, the information processing method, and the recording medium will be described. Hereinafter, the third embodiment of the information processing device, the information processing method, and the recording medium will be described using a detection device 3 to which the third embodiment of the information processing device, the information processing method, and the recording medium is applied. [3-1: Configuration of the detection device 3]
[0056] The configuration of the detection device 3 in the third embodiment will be described with reference to Fig. 5. Fig. 5 is a block diagram showing the configuration of the detection device 3 in the third embodiment.
[0057] As shown in FIG. 5 , the detection device 3 in the third embodiment includes a calculation device 21 and a storage device 22, similar to the learning device 2 in the second embodiment. Furthermore, the detection device 3 in the third embodiment may include a communication device 23, an input device 24, and an output device 25, similar to the learning device 2 in the second embodiment. However, the detection device 3 does not necessarily have to include at least one of the communication device 23, the input device 24, and the output device 25. The detection device 3 in the third embodiment differs from the learning device 2 in the second embodiment in that it performs a detection operation using the detection mechanism DM learned by the learning device 2 in the second embodiment. Other features of the detection device 3 may be the same as those of the learning device 2. Therefore, the following will describe in detail the differences from the previously described embodiments, and will omit appropriate descriptions of other overlapping features. [3-2: Detection Operation Performed by the Detection Device 3]
[0058] The flow of the detection operation performed by the detection device 3 will be described with reference to Fig. 6. Fig. 6 is a flowchart showing the detection operation performed by the detection device 3. In the third embodiment, an example will be described in which facial feature points are detected from a face region of an image. The parameter storage unit 322 in the third embodiment stores the parameters of the detection mechanism DM learned in the second embodiment.
[0059] 6 , the acquisition unit 311 acquires an image (step S30). The face detection unit 316 detects a face region from the image (step S31). The feature detection unit 313 detects feature points from the face region using the detection mechanism DM (step S32). The feature points detected from the face region may be applied to face authentication, facial expression analysis, etc.
[0060] If the input image is a human body image, the detection mechanism DM may output parts such as elbows, knees, and heels as feature points. [3-3: Technical Effects of the Detection Device 3]
[0061] The detection mechanism DM uses a computationally efficient algorithm and a convolutional neural network, allowing for high-speed calculations. The detection device 3 in the third embodiment uses the detection mechanism DM to detect feature information from an image, allowing for accurate and high-speed detection of feature information. [4: Fourth Embodiment]
[0062] Next, a fourth embodiment of an information processing device, an information processing method, and a recording medium will be described. Hereinafter, the fourth embodiment of the information processing device, the information processing method, and the recording medium will be described using an authentication device 4 to which the fourth embodiment of the information processing device, the information processing method, and the recording medium is applied. [4-1: Configuration of authentication device 4]
[0063] The configuration of the authentication device 4 in the fourth embodiment will be described with reference to Fig. 7. Fig. 7 is a block diagram showing the configuration of the authentication device 4 in the fourth embodiment.
[0064] As shown in FIG. 7 , the authentication device 4 of the fourth embodiment includes a computing device 21 and a storage device 22, similar to the learning device 2 of the second embodiment and the detection device 3 of the third embodiment. Furthermore, the authentication device 4 of the fourth embodiment may include a communication device 23, an input device 24, and an output device 25, similar to the learning device 2 of the second embodiment and the detection device 3 of the third embodiment. However, the authentication device 4 does not necessarily have to include at least one of the communication device 23, the input device 24, and the output device 25. The authentication device 4 of the fourth embodiment differs from the learning device 2 of the second embodiment and the detection device 3 of the third embodiment in that it performs authentication operations using the detection mechanism DM learned by the learning device 2 of the second embodiment. Other features of the authentication device 4 may be the same as those of at least one of the learning device 2 and the detection device 3. Therefore, the following will describe in detail the differences from the previously described embodiments, and will omit appropriate descriptions of other overlapping features. [4-2: Detection Operation Performed by the Authentication Device 4]
[0065] The flow of the detection operation performed by the authentication device 4 will be described with reference to Fig. 8. Fig. 8 is a flowchart showing the detection operation performed by the authentication device 4. In the fourth embodiment, an example will be described in which face authentication is performed using a face region of an image. The parameter storage unit 322 in the fourth embodiment stores the parameters of the detection mechanism DM learned in the second embodiment.
[0066] As shown in FIG. 8, the acquisition unit 311 acquires an image (step S30). The face detection unit 316 detects a face area from the image (step S31). The feature detection unit 313 detects feature points from the face area using the detection mechanism DM (step S32). The normalization unit 417 normalizes the face orientation in the face image to a predetermined orientation and the face size to a predetermined size (step S40). The extraction unit 418 extracts feature amounts from the normalized face (step S41). The extraction unit 418 extracts feature amounts useful for face authentication from the face image. The authentication unit 419 compares and collates the feature amounts extracted in step S41 with feature amounts registered in the registration information storage unit 423 to determine whether they are the same person (step S42). [4-3: Technical Effects of Authentication Device 4]
[0067] The detection mechanism DM can detect feature points from a face image with high accuracy and high speed. Therefore, the authentication device 4 in the fourth embodiment performs face authentication based on feature points detected from a face image using the detection mechanism DM, and therefore can perform face authentication with high accuracy and high speed. [5: Supplementary Note]
[0068] The following supplementary notes are further disclosed with respect to the above-described embodiments. [Supplementary Note 1] An information processing device comprising: an acquisition means for acquiring a supervised image having supervised feature information and a supervised non-image not having supervised feature information; and a learning means for causing a detection mechanism to learn a detection method for detecting feature information from an input image, using the supervised feature information detected by the detection mechanism from the supervised image and supervised detection information based on the supervised feature information, as well as the supervised non-image detected by the detection mechanism and supervised non-detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the supervised non-image. [Supplementary Note 2] The information processing device according to Supplementary Note 1, comprising: the supervised detection information is information based on a supervised loss function in which a loss increases as the similarity between the supervised feature information and the supervised feature information decreases; and the supervised non-detection information is information based on a supervised loss function in which a loss increases as the similarity between the supervised non-feature information and the extended feature information decreases. [Supplementary Note 3] The information processing device according to Supplementary Note 1 or 2, comprising: a transforming means for applying a transform to the non-correct image to generate the extended image by extending the non-correct image; and an inverse transforming means for applying an inverse transform to the extended feature information detected from the extended image by the detection mechanism. [Supplementary Note 4] The information processing device according to any one of Supplements 1 to 3, wherein the acquiring means acquires more non-correct images compared to the correct images. [Supplementary Note 5] The information processing device according to any one of Supplements 1 to 4, comprising: a detecting means for detecting feature information from an image using the detection mechanism. [Supplementary Note 6] The information processing device according to any one of Supplements 1 to 5, comprising: an extracting means for extracting a feature amount from the image using the feature information; and an authenticating means for authenticating the image using the feature amount. [Supplementary Note 7] The information processing device according to any one of Supplements 1 to 6, wherein the image is a facial image. [Supplementary Note 8] The information processing device according to any one of Supplementary Notes 1 to 7, wherein the feature information is a feature point of the image.[Supplementary Note 9] An information processing method comprising: acquiring a supervised image having supervised feature information and a supervised non-image having no supervised feature information; and causing a detection mechanism to learn a detection method for detecting feature information from an input image using the supervised feature information detected by the detection mechanism from the supervised image and supervised detection information based on the supervised feature information, as well as supervised non-feature information detected by the detection mechanism from the supervised non-image and supervised non-detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the supervised non-image. [Supplementary Note 10] A recording medium having recorded thereon a computer program for causing a computer to execute an information processing method for acquiring a correct-answer image having correct correct-answer feature information and a correct-answer non-image having no correct answer feature information, and causing the detection mechanism to learn a detection method for detecting feature information from an input image using the correct-answer feature information detected by the detection mechanism from the correct-answer image and correct-answer detection information based on the correct answer feature information, as well as correct-answer featureless information detected by the detection mechanism from the correct-answer non-image and correct-answer non-detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the correct-answer non-image.
[0069] This disclosure may be modified as appropriate within the scope of the claims and the technical idea that can be read from the entire specification. Information processing devices, information processing methods, and recording media that involve such modifications are also included in the technical idea of this disclosure.
[0070] REFERENCE SIGNS LIST 1 Information processing device 11, 211, 311 Acquisition unit 12, 212 Learning unit 2 Learning device 2121 Loss function calculation unit 2122 Parameter update unit DM Detection mechanism 213 Detection unit 214 Transformation unit 215 Inverse transformation unit 221 Learning data storage unit 222, 322 Parameter storage unit 3 Detection device 313 Feature detection unit 316 Face detection unit 4 Authentication device 417 Normalization unit 418 Extraction unit 419 Authentication unit 423 Registration information storage unit
Claims
1. an acquisition means for acquiring a correct image having correct feature information and a correct image not having correct feature information; a learning means for causing the detection mechanism to learn a detection method for detecting feature information from an input image, using correct-answer feature information detected by the detection mechanism from the correct-answer image, correct-answer detection information based on the correct-answer feature information, correct-answer no-answer information detected by the detection mechanism from the correct-answer no-answer image, and correct-answer no-answer detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the correct-answer no-answer image; An information processing device comprising:
2. the correct answer presence detection information is information based on a correct answer presence loss function in which the loss increases as the correct answer presence feature information and the correct answer feature information are less similar to each other, The correct answer no-detection information is information based on a correct answer lossless function in which the loss increases as the correct answer no-feature information and the extended feature information are less similar. The information processing device according to claim 1 , comprising:
3. a conversion means for converting the correct non-image to generate the expanded image by expanding the correct non-image; an inverse transformation means for performing an inverse transformation of the transformation on the extended feature information detected from the extended image by the detection mechanism; The information processing device according to claim 1 or 2, comprising:
4. The acquiring means acquires a larger number of the non-correct images compared to the correct images.
3. The information processing device according to claim 1.
5. A detection means for detecting feature information from an image using the detection mechanism. The information processing device according to claim 1 or 2, comprising:
6. an extraction means for extracting a feature amount from the image using the feature information; authentication means for authenticating the image using the feature amount; The information processing device according to claim 1 or 2, comprising:
7. The image is a face image.
3. The information processing device according to claim 1.
8. The feature information is a feature point of the image.
3. The information processing device according to claim 1.
9. Acquire correct images having correct feature information and correct images not having correct feature information; The detection mechanism is made to learn a detection method for detecting feature information from an input image using correct-answer feature information detected by the detection mechanism from the correct-answer image, correct-answer detection information based on the correct-answer feature information, correct-answer no-feature information detected by the detection mechanism from the correct-answer no-image, and correct-answer no-detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the correct-answer no-image. A computer-implemented information processing method.
10. On the computer, Acquire correct images having correct feature information and correct images not having correct feature information; The detection mechanism is made to learn a detection method for detecting feature information from an input image using correct-answer feature information detected by the detection mechanism from the correct-answer image, correct-answer detection information based on the correct-answer feature information, correct-answer no-feature information detected by the detection mechanism from the correct-answer no-image, and correct-answer no-detection information based on extended feature information detected by the detection mechanism from an extended image obtained by extending the correct-answer no-image. A computer program for executing an information processing method.