Face verification methods, devices, equipment and computer-readable storage media

Abstract

This disclosure provides a face verification method, apparatus, device, and computer-readable storage medium, relating to artificial intelligence technology and image processing technology. The solution includes: acquiring original images; extracting facial feature information from the original images using a feature extraction network; inputting the facial feature information into a keypoint recognition network and a liveness detection network, wherein the keypoint recognition network is used to determine keypoint information, and the liveness detection network is used to determine liveness detection information; and determining the face verification result based on the keypoint information and liveness detection information of multiple frames of original images. In the face verification method, apparatus, device, and computer-readable storage medium provided in this application, face detection and keypoint detection share a single set of facial feature information. Therefore, in the face verification process, facial feature information only needs to be extracted once, thereby reducing the data processing volume of the face verification process and improving the face verification speed.

Description

Technical Field

[0001] This disclosure relates to artificial intelligence technology and image processing technology, and in particular to a face verification method, apparatus, device and computer-readable storage medium. Background Technology

[0002] Currently, facial verification technology is used in many application scenarios. For example, when logging into a payment app, facial verification can be used. If the detected face matches a preset face, the verification is successful.

[0003] In addition to facial verification, liveness detection is also required. For example, liveness detection can be performed by using commands and actions, such as turning the face left or right. If the commands are not executed correctly, the verification is considered to have failed.

[0004] Because face detection requires both face detection and liveness detection, the face verification process involves a large amount of data processing and is slow. Summary of the Invention

[0005] This disclosure provides a face verification method, apparatus, device, and computer-readable storage medium to solve the problem of slow face verification speed in the prior art.

[0006] The first aspect of this disclosure is to provide a face verification method, including:

[0007] Acquire the original image and extract facial feature information from the original image using a feature extraction network;

[0008] The facial feature information is input into a key point recognition network and a liveness detection network, respectively, wherein the key point recognition network is used to determine key point information and the liveness detection network is used to determine liveness detection information;

[0009] The face verification result is determined based on the key point information and the liveness detection information of the multiple frames of the original image.

[0010] The first aspect of this disclosure is to provide a face verification device, comprising:

[0011] Acquire the original image and extract facial feature information from the original image using a feature extraction network;

[0012] The facial feature information is input into a key point recognition network and a liveness detection network, respectively, wherein the key point recognition network is used to determine key point information, and the liveness detection network is used to determine;

[0013] The face verification result is determined based on the key point information and the liveness detection information of the multiple frames of the original image.

[0014] Another aspect of this disclosure is to provide a face verification device, comprising:

[0015] Memory;

[0016] Processor; and

[0017] Computer programs;

[0018] The computer program is stored in the memory and configured to be executed by the processor to implement the face verification method as described in the first aspect above.

[0019] Another aspect of this disclosure is to provide a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the face verification method as described in the first aspect above.

[0020] The technical advantages of the face verification method, apparatus, device, and computer-readable storage medium disclosed herein are:

[0021] The face verification method, apparatus, device, and computer-readable storage medium disclosed herein include: acquiring an original image; extracting facial feature information from the original image using a feature extraction network; inputting the facial feature information into a keypoint recognition network and a liveness detection network, wherein the keypoint recognition network is used to determine keypoint information, and the liveness detection network is used to determine liveness detection information; and determining the face verification result based on the keypoint information and liveness detection information of multiple frames of original images. In the face verification method, apparatus, device, and computer-readable storage medium provided herein, face detection and keypoint detection share a single set of facial feature information. Therefore, during the face verification process, facial feature information only needs to be extracted once, thereby reducing the data processing volume of the face verification process and improving the face verification speed. Attached Figure Description

[0022] Figure 1 An application scenario diagram illustrating an exemplary embodiment of this application;

[0023] Figure 2 A flowchart illustrating a face verification method as shown in an exemplary embodiment of this application;

[0024] Figure 3 A flowchart illustrating a face verification method as another exemplary embodiment of the present invention;

[0025] Figure 4A This is a schematic diagram illustrating the determination of a facial image in an original image, as shown in an exemplary embodiment of this application.

[0026] Figure 4B The processing procedure of the original image is shown as an exemplary embodiment of this application;

[0027] Figure 5 This is a structural diagram of a face verification device shown in an exemplary embodiment of the present invention;

[0028] Figure 6 A structural diagram of a face verification device shown in another exemplary embodiment of the present invention;

[0029] Figure 7 This is a structural diagram of a face verification device illustrated in an exemplary embodiment of the present invention. Detailed Implementation

[0030] Figure 1 This is an application scenario diagram illustrating an exemplary embodiment of this application.

[0031] like Figure 1 As shown, during the face verification process, the camera 11 of the electronic device captures multiple consecutive frames of user images and extracts the facial information from the images. The facial information is then compared with the preset facial information. If the two match, it can be determined that the captured user images include the preset face.

[0032] To prevent malicious verification using photos or images of pre-set faces, a liveness detection step has been added to the face verification process.

[0033] Liveness detection methods include using commands to coordinate actions, such as turning the face left or right, opening the mouth, or blinking. If the commands are not coordinated correctly, the face verification is considered to have failed.

[0034] Liveness detection methods can also include detection methods based on TOF (Time of Flight) technology. This method can reconstruct the face and obtain point cloud maps and depth maps of the face and background, thus serving as accurate liveness detection.

[0035] However, in the face verification process, face comparison and liveness detection are performed independently, which means that the terminal needs to perform both face comparison and liveness detection calculations at the same time. This places high demands on the terminal's computing power, and the fact that the terminal needs to perform two calculations when performing face verification results in a slow verification speed.

[0036] Figure 2 This is a flowchart illustrating a face verification method as an exemplary embodiment of this application.

[0037] like Figure 2 As shown, the face verification method provided in this application includes:

[0038] Step 201: Acquire the original image and extract facial feature information from the original image using a feature extraction network.

[0039] The face verification method provided in this application can be executed by an electronic device with computing power, such as a tablet computer or a smartphone.

[0040] The electronic device can be equipped with an image acquisition device, such as a camera, to capture raw images. The electronic device can also communicate with the image acquisition device, allowing it to capture raw images and send them back to the electronic device.

[0041] Specifically, when facial verification is required, the electronic device can execute the method provided in this application. The method provided in this application can be encapsulated in software, and the software can also be configured in the electronic device. When the electronic device runs the software, it can execute the facial verification method provided in this application.

[0042] Furthermore, when facial verification is required, the camera of the electronic device can be turned on, thereby capturing the original image through the camera.

[0043] In practical applications, a feature extraction network can be pre-configured to extract facial feature information from the original image. This feature extraction network can be a neural network structure, and its output feature map can be used as facial feature information.

[0044] The feature extraction network can be pre-trained. For example, a neural network can be built, or a large number of images can be collected and facial features can be labeled in them. The labeled data can then be used to train the neural network to obtain the feature extraction network.

[0045] Specifically, a face detection network can be set up to process the original image and identify facial features. For example, the face detection network can determine a bounding box in the original image, and the area selected by the bounding box is the face.

[0046] Furthermore, the face detection network can be pre-trained. For example, a neural network can be built, and a large number of images can be collected and face regions can be labeled in them. The labeled data can then be used to train the neural network to obtain the face detection network.

[0047] Step 202: Input the facial feature information into the key point recognition network and the liveness detection network respectively. The key point recognition network is used to determine key point information, and the liveness detection network is used to determine liveness detection information.

[0048] In practical applications, the facial feature information extracted by the feature network can be input into the keypoint recognition network and the liveness detection network, respectively. This allows the keypoint recognition network to output keypoint information and the liveness detection network to output liveness detection information.

[0049] In this implementation, the key point recognition network for key point detection and the liveness detection network for liveness detection can share a set of facial feature information. Thus, during the face verification process, facial feature information only needs to be determined once for an original image, thereby reducing the amount of data processing, reducing the data processing pressure on the terminal, and improving the face verification speed.

[0050] Among them, a key point recognition network can be pre-set. The key point information determined by the network can locate key facial information, such as the location information of key points such as the tip of the nose, the root of the nose, the chin, the outer corner of the left eye, the inner corner of the left eye, the outer corner of the right eye, the inner corner of the right eye, the center of the mouth, the left corner of the mouth, the right corner of the mouth, the outermost part of the left face, and the outermost part of the right face.

[0051] Specifically, the key point recognition network can be pre-trained. For example, a neural network can be built, or a large number of images can be collected and key points can be labeled in them. The labeled data can then be used to train the neural network to obtain the key point recognition network.

[0052] Furthermore, a liveness detection network can be pre-configured to determine whether the original image contains a real human face. For example, if the image does not contain a real human face, the output of the liveness detection network can be 0; if the image contains a real human face, the output of the liveness detection network can be 1.

[0053] In practical applications, the liveness detection network can be pre-trained. For example, a binary classification model can be built, and a large number of images can be collected. These images can be continuous images, such as images obtained by recording real faces or images obtained by recording facial photographs. These images can be labeled, and then the binary classification model can be trained using these labeled data to obtain the liveness detection network.

[0054] Step 203: Determine the face verification result based on the key point information and liveness detection information of multiple original images.

[0055] When performing facial verification, electronic devices can capture multiple consecutive frames of raw images. For example, if the camera of an electronic device can be turned on for n seconds, and the electronic device captures p frames of raw images per second, then the electronic device can capture a total of n*p raw images.

[0056] Specifically, for each original image, corresponding key point information and liveness detection information can be determined. Therefore, the face verification result can be determined by combining the key point information and liveness detection information from multiple original images.

[0057] Furthermore, key point information can be used to determine whether the face in the original image matches a preset face. For example, preset key point information can be extracted from the preset face. After extracting key point information from the original image, the extracted key point information can be compared with the preset key point information. If they match, then the face in the original image can be considered to match the preset face.

[0058] In practical applications, key point information can locate key points on the face. Based on key points in multiple frames of original images, it can be determined whether the original image contains a real human face. For example, based on key point information from multiple frames of original images, it can be determined whether behaviors such as blinking, opening the mouth, closing the mouth, or shaking the head exist. If they do, then it can be assumed that a real human face exists in the original image.

[0059] If the liveness detection network determines that a face exists in the original image based on facial feature information, it can also be considered that a real face exists in the original image.

[0060] Specifically, the face verification result can be determined by combining key point information and liveness detection information. For example, if the key point information determines that there is a real face in the original image, and the liveness detection information also determines that there is a real face in the original image, then it can be considered that there is a real face in the original image.

[0061] Furthermore, if the face in the original image matches the preset face based on key point information and liveness detection information, and the original image contains a real face, then the face verification can be considered successful; otherwise, the face verification is considered unsuccessful.

[0062] The method provided in this embodiment is used for face verification. The method is executed by a device equipped with the method provided in this embodiment, which is typically implemented in hardware and / or software.

[0063] The face verification method provided in this embodiment includes: acquiring an original image; extracting facial feature information from the original image using a feature extraction network; inputting the facial feature information into a keypoint recognition network and a liveness detection network, respectively, wherein the keypoint recognition network is used to determine keypoint information, and the liveness detection network is used to determine liveness detection information; and determining the face verification result based on the keypoint information and liveness detection information of multiple frames of original images. In the face verification method provided in this application, face detection and keypoint detection share a set of facial feature information. Therefore, in the face verification process, facial feature information only needs to be extracted once, thereby reducing the data processing volume of the face verification process and improving the face verification speed.

[0064] Figure 3 A flowchart illustrating a face verification method as another exemplary embodiment of the present invention.

[0065] like Figure 3 As shown, the face verification method provided in this embodiment includes:

[0066] Step 301: Acquire the original image and determine the facial image from the original image.

[0067] The method of acquiring the original image is the same as the execution method and principle of step 201, and will not be repeated here.

[0068] Specifically, facial images can be identified from the raw images captured by electronic devices. For example, a face detection network can be pre-set to process the raw images and identify facial images within them. For instance, the face detection network can determine bounding boxes in the raw images, and these bounding boxes define the areas representing the faces.

[0069] Furthermore, the original image can be input into the face detection network, allowing the network to identify facial images from the original image.

[0070] Figure 4A This is a schematic diagram illustrating the determination of a facial image in an original image, as shown in an exemplary embodiment of this application.

[0071] like Figure 4A As shown, the original image 41 can be input into the face detection network, which can process the original image 41 to determine a rectangular box 42 in the original image 41. The portion selected by the rectangular box 42 is the facial image in the original image.

[0072] Step 302: Input the facial image into the feature extraction network and determine the facial feature information of the facial image through the feature extraction network.

[0073] In practical applications, the facial image determined in the original image can be input into the feature extraction network, thereby extracting the facial feature information of the facial image through the feature extraction network.

[0074] The first feature map and the second feature map are used as inputs to the liveness detection network; the second feature map is used as inputs to the key point recognition network.

[0075] Specifically, the feature extraction network can include convolutional layers and pooling layers. Convolutional layers extract features from facial images, and pooling layers output feature maps of the facial images.

[0076] Furthermore, the feature extraction network includes a regular convolutional layer, several separable convolutional layers, a first pooling layer, and a second pooling layer; the first pooling layer outputs a first feature map, and the second pooling layer outputs a second feature map.

[0077] In practical applications, facial images can be resized to a preset size before being input into the feature extraction network. For example, a facial image can be resized to (256, 256, 3) before being input into the feature extraction network.

[0078] In this feature extraction network, the first convolutional layer is a regular convolutional layer, while the other convolutional layers are separable convolutional layers, thereby reducing the size of the network and the amount of computation.

[0079] Specifically, the first pooling layer can be a max pooling layer. The feature extraction network may also include a first max pooling layer and a first average pooling layer. After processing the input image through a regular convolutional layer, a first max pooling layer, a portion of separable convolutional layers, and a first average pooling layer, a feature map of size (128, 128, 128) can be obtained. This (128, 128, 128) feature map can be input into the first pooling layer to obtain a first feature map of size (32, 32, 128).

[0080] The second pooling layer can also be a max pooling layer. The feature extraction network can also include a second max pooling layer and a second average pooling layer. The first feature map can be processed by the second max pooling layer, the remaining separable convolutional layers and the second average pooling layer to obtain a feature map of size (64, 64, 160). This (64, 64, 160) feature map can be input into the second pooling layer to obtain a second feature map of size (32, 32, 160).

[0081] Furthermore, a first feature map of size (32, 32, 128) and a second feature map of size (32, 32, 160) can be input into the liveness detection network to obtain the liveness detection result.

[0082] In practical applications, a second feature map of size (32, 32, 160) can be input into the key point recognition network to obtain key point information.

[0083] The key point information includes the location of key points. The key point recognition network determines the location of key points based on facial feature information, specifically based on the second feature map. Examples include the location of the eyes, mouth, nose, eyebrows, and the edges of the face.

[0084] Step 303: Input the facial feature information into the key point recognition network and the liveness detection network respectively. The key point recognition network is used to determine key point information, and the liveness detection network is used to determine liveness detection information.

[0085] The specific principles and implementation methods of step 303 are similar to those of step 202, and will not be repeated here.

[0086] Figure 4B The processing procedure of the original image is shown as an exemplary embodiment of this application.

[0087] like Figure 4B As shown, the acquired raw image 43 can be input into the face detection network 44, which can identify the facial image. The facial image is then input into the feature extraction network 45, which can identify the first feature map and the second feature map. The first and second feature maps are then input into the liveness detection network 46, and the second feature map is input into the keypoint recognition network 47. The liveness detection network 46 can determine liveness detection information, and the keypoint recognition network 47 can determine keypoint information.

[0088] Step 304: Determine the face comparison result based on the key point information of multiple original images and the preset face information.

[0089] Specifically, key point information from each original image can be used to determine whether the face in the original image is a preset face.

[0090] Furthermore, the preset face can be recorded by the user when registering an account. The electronic device or backend server can record this preset face and determine its information. For example, the preset face information can be extracted from the preset face using the feature extraction network described in this application.

[0091] In practical applications, when electronic devices perform face verification, they can determine whether the face in the original image matches the preset face based on the key point information of multiple original images and the preset face information; if they match, the face comparison result is determined to be consistent.

[0092] This process involves acquiring preset facial information and comparing it with key point information determined from multiple frames of original images. If the comparison results match, the facial comparison is considered consistent. If the comparison results do not match, the facial comparison is considered inconsistent.

[0093] For example, a preset ratio can be set. If the key point information of the image that reaches the preset ratio in multiple original images is consistent with the preset face information, then the comparison result can be considered consistent.

[0094] Step 305: Determine the real face result based on the key point information and liveness detection information of multiple frames of original images.

[0095] Furthermore, the electronic device can determine whether the original image contains a real human face based on key point information and liveness detection information from multiple frames of original images.

[0096] Specifically, key point information includes the location of key points on the face. Based on the location of key points in multiple frames of images, it can be determined whether there is any facial movement, such as blinking or opening the mouth.

[0097] Furthermore, the presence of real human faces in the original images can be determined based on liveness detection information. For example, if the liveness detection information of an original image is 1, it can be assumed that the original image contains a real human face; if the liveness detection information is 0, it can be assumed that the original image does not contain a real human face. If it is determined that real human faces exist in most of the original images, then it can be determined that real human faces exist in multiple frames of original images.

[0098] In practical applications, the liveness detection information output by the liveness detection network includes classification results, such as whether real faces are included or not.

[0099] Specifically, when determining the real face result based on the key point information and liveness detection information of multiple original images, the key point activity information can be determined based on the key point information of multiple original images; then, the real face result can be determined based on the classification results and key point activity information in the liveness detection information.

[0100] Specifically, key point information includes the location of facial key points, such as the position of facial features. Based on this key point information, the activity information of key points in multiple original images can be determined.

[0101] Furthermore, the presence of facial feature activity can be determined based on the positional information of facial features in multiple frames of original images. For example, the position of the eyes in multiple frames of original images can determine whether blinking has occurred. The position of the mouth in multiple frames of original images can determine whether the mouth is opening or closing. Furthermore, the position of key points in the keypoint information can determine whether head shaking has occurred.

[0102] Furthermore, keypoint activity information can characterize whether a real human face is included in multiple original images, and the classification results in the liveness detection information can also characterize whether a real human face is included in multiple original images. Therefore, combining these two pieces of information can determine whether a real human face is included in multiple original images, making the determination more accurate.

[0103] In practical applications, if the classification result is a real face result, and / or the key point activity information shows that there is facial feature activity, then it is determined that the original image includes a real face.

[0104] Step 306: Determine the face verification result based on the face comparison result and the real face result.

[0105] In practical applications, if the face comparison result is consistent with the original image and the real face result is that the original image contains a real face, then the face verification is deemed successful. Otherwise, the face verification is deemed unsuccessful.

[0106] If the faces in the multiple original images match the preset faces, it means that the original images collected include the user's own facial image. If real faces are present in the multiple original images, it can be considered that a real person is performing face verification. Therefore, if the face comparison results are consistent and the original images include real faces, the face verification can be considered successful.

[0107] Figure 5 This is a structural diagram of a face verification device shown in an exemplary embodiment of the present invention.

[0108] like Figure 5 As shown, the face verification device provided in this embodiment includes:

[0109] Feature extraction module 51 is used to acquire the original image and extract facial feature information from the original image through a feature extraction network;

[0110] The recognition module 52 is used to input the facial feature information into the key point recognition network and the liveness detection network respectively, wherein the key point recognition network is used to determine key point information and the liveness detection network is used to determine liveness detection information;

[0111] The verification module 53 is used to determine the face verification result based on the key point information and the liveness detection information of the multiple frames of the original images.

[0112] The face verification device provided in this application includes a feature extraction module for acquiring original images and extracting facial feature information from the original images through a feature extraction network; a recognition module for inputting the facial feature information into a keypoint recognition network and a liveness detection network, wherein the keypoint recognition network is used to determine keypoint information and the liveness detection network is used to determine liveness detection information; and a verification module for determining the face verification result based on the keypoint information and liveness detection information of multiple frames of original images. In the face verification device provided in this application, face detection and keypoint detection share a single set of facial feature information. Therefore, during the face verification process, facial feature information only needs to be extracted once, thereby reducing the data processing volume of the face verification process and improving the face verification speed.

[0113] The specific principles and implementation methods of the face verification device provided in this embodiment are all the same as those in this embodiment. Figure 2 The embodiments shown are similar and will not be described again here.

[0114] Figure 6This is a structural diagram of a face verification device shown as another exemplary embodiment of the present invention.

[0115] like Figure 6 As shown, based on the above embodiments, the face verification device provided in this embodiment optionally includes the feature extraction module 51, which comprises:

[0116] Image determination unit 511 is used to determine a facial image in the original image;

[0117] The feature extraction unit 512 is used to input the facial image into the feature extraction network and determine the facial feature information of the facial image through the feature extraction network.

[0118] Optionally, the key point recognition network determines the location of key points based on the facial feature information.

[0119] Optionally, the verification module 53 includes:

[0120] The face comparison unit 531 is used to determine the face comparison result based on the key point information and preset face information of the multiple frames of the original images;

[0121] The real face recognition unit 532 is used to determine the real face result based on the key point information and the liveness detection information of the multiple frames of the original images;

[0122] Verification unit 533 is used to determine the face verification d based on the face comparison result and the real face result.

[0123] Optionally, the liveness detection information includes classification results;

[0124] The real face recognition unit 532 is specifically used for:

[0125] Key point activity information is determined based on the key point information of the original images in multiple frames;

[0126] The real face result is determined based on the classification result in the liveness detection information and the key point activity information.

[0127] Optionally, the key point information includes facial feature location information;

[0128] The real face recognition unit 532 is specifically used for:

[0129] Based on the facial feature position information of the original images in multiple frames, determine whether there is any facial feature activity.

[0130] Optionally, the real face recognition unit 532 is specifically used for:

[0131] If the classification result is a real face result, and / or the key point activity information is a result showing facial feature activity, then it is determined that the original image includes a real face.

[0132] Optionally, the face comparison unit 531 is specifically used for:

[0133] Based on the key point information and preset face information of the original images in multiple frames, determine whether the face in the original image is consistent with the preset face;

[0134] If they match, then the face comparison results are considered to be consistent.

[0135] Optionally, the verification unit 533 is specifically used for:

[0136] If the face comparison result is consistent with the face comparison result, and the real face result is that the original image includes a real face, then the face verification is determined to be successful.

[0137] Optionally, the feature extraction network determines a first feature map and a second feature map based on the facial image;

[0138] The first feature map and the second feature map are used as inputs to the liveness detection network; the second feature map is used as inputs to the key point recognition network.

[0139] Optionally, the feature extraction network includes a conventional convolutional layer, several separable convolutional layers, a first pooling layer, and a second pooling layer;

[0140] The first pooling layer outputs the first feature map, and the second pooling layer outputs the second feature map.

[0141] The specific principles and implementation methods of the face verification device provided in this embodiment are all the same as those in this embodiment. Figure 3 The embodiments shown are similar and will not be described again here.

[0142] Figure 7 This is a structural diagram of a face verification device illustrated in an exemplary embodiment of the present invention.

[0143] like Figure 7 As shown, the face verification device provided in this embodiment includes:

[0144] Memory 71;

[0145] Processor 72; and

[0146] Computer programs;

[0147] The computer program is stored in the memory 71 and configured to be executed by the processor 72 to implement any of the face verification methods described above.

[0148] This embodiment also provides a computer-readable storage medium on which a computer program is stored.

[0149] The computer program is executed by a processor to implement any of the face verification methods described above.

[0150] This embodiment also provides a computer program, including program code, which, when the computer runs the computer program, executes any of the face verification methods described above.

[0151] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A face verification method, characterized in that, The method comprises the following steps: Collecting multiple frames of original images, and extracting face feature information in the original images through a feature extraction network; Inputting the face feature information into a key point recognition network and a living body detection network respectively, wherein the key point recognition network is used to determine key point information, and the living body detection network is used to determine living body detection information, and the living body detection network is a pre-trained binary classification model; Determining a face verification result according to the key point information and the living body detection information of the multiple frames of original images; The face feature information in the original images is extracted through the feature extraction network, which comprises the following steps: Determining a face image in the original image, and adjusting the face image to a preset size; Inputting the adjusted face image into the feature extraction network, and the feature extraction network processes and outputs a first feature map and a second feature map in parallel according to the face image; The first feature map and the second feature map are used to input the living body detection network to determine the living body detection information, and the second feature map is used to input the key point recognition network to determine the key point information; The feature extraction network comprises a regular convolution layer, a plurality of separable convolution layers, a first pooling layer and a second pooling layer; The first pooling layer outputs the first feature map, and the second pooling layer outputs the second feature map; The face verification result is determined according to the key point information and the living body detection information of the multiple frames of original images, which comprises the following steps: Determining a face comparison result according to the key point information and preset face information of the multiple frames of original images; Determining a real face result according to the key point information and the living body detection information of the multiple frames of original images; Determining the face verification result according to the face comparison result and the real face result; The living body detection information comprises a classification result; The real face result is determined according to the key point information and the classification result in the living body detection information, which comprises the following steps: Determining key point activity information according to the key point information of the multiple frames of original images; Determining the real face result according to the classification result in the living body detection information and the key point activity information; The key point information comprises five feature position information; The key point activity information is determined according to the five feature position information of the multiple frames of original images, which comprises the following steps: Determining whether there is five feature activity result according to the five feature position information of the multiple frames of original images; If the classification result is a real face result and / or the key point activity information is a five feature activity result, it is determined that the original image comprises a real face.

2. The method of claim 1, wherein, The key point recognition network determines key point positions according to the face feature information.

3. The method of claim 1, wherein, The face comparison result is determined according to the key point information and preset face information of the multiple frames of original images, which comprises the following steps: Determining whether the face in the original image is consistent with the preset face according to the key point information and preset face information of the multiple frames of original images; If they are consistent, it is determined that the face comparison result is consistent.

4. The method of claim 1, wherein, The face verification result is determined according to the face comparison result and the real face result. If the face comparison result is consistent and the real face result is that the original image includes a real face, it is determined that the face verification is passed.

5. A face authentication apparatus characterized by comprising: Comprise: A feature extraction module is configured to collect an original image and extract face feature information in the original image through a feature extraction network; An identification module is configured to input the face feature information into a key point identification network and a living body detection network, respectively, wherein the key point identification network is configured to determine key point information, and the living body detection network is configured to determine living body detection information, and the living body detection network is a pre-trained binary classification model; A verification module is configured to determine a face verification result according to the key point information and the living body detection information of multiple frames of the original image, and the verification module comprises a face comparison unit, a real face identification unit, and a verification unit; The feature extraction module is specifically configured to determine a face image in the original image, adjust the face image to a preset size, input the adjusted face image into the feature extraction network, and the feature extraction network processes and outputs a first feature map and a second feature map in parallel according to the face image; wherein the first feature map and the second feature map are collectively used to input the living body detection network to determine the living body detection information; the second feature map is used to input the key point identification network to determine the key point information; the feature extraction network comprises a regular convolution layer, a plurality of separable convolution layers, a first pooling layer, and a second pooling layer; the first pooling layer outputs the first feature map, and the second pooling layer outputs the second feature map; The face comparison unit is configured to determine a face comparison result according to the key point information of multiple frames of the original image and preset face information; The real face identification unit is configured to determine a real face result according to the key point information and the living body detection information of multiple frames of the original image; The verification unit is configured to determine the face verification result according to the face comparison result and the real face result; The living body detection information comprises a classification result; the real face identification unit is specifically configured to determine key point activity information according to the key point information of multiple frames of the original image; and determine the real face result according to the classification result in the living body detection information and the key point activity information; The key point information comprises five facial feature position information; the real face identification unit is specifically configured to determine whether there is a five facial feature activity result according to the five facial feature position information of multiple frames of the original image; if the classification result is a real face result and / or the key point activity information is a five facial feature activity result, it is determined that the original image includes a real face.

6. A face device, characterized by Comprise: A memory; A processor; And A computer program; The computer program is stored in the memory and configured to be executed by the processor to implement the method of any one of claims 1-4.

7. A computer readable storage medium characterized in that, A computer program product comprising a computer readable medium having stored thereon a computer program, the computer program being executable by a processor to implement the method of any one of claims 1-4.

8. A computer program product, characterised in that, A computer program product comprising a computer readable medium having stored thereon a computer program, the computer program being executable by a processor to implement the method of any one of claims 1-4.

Images