Facial feature updating method, device, computer equipment and storage medium

Abstract

The present application relates to a facial feature update method, device, computer equipment and storage medium. The method includes: determining a facial image base library, a facial image test library, and a facial image update set; training a facial recognition model based on the facial image test library to obtain an error recognition rate; when the error recognition rate meets the facial feature update condition, extracting facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set, respectively, and updating the features of the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features. The method can quickly further reduce the false alarm rate and thus improve the accuracy rate without retraining the original facial recognition model while ensuring the accuracy rate of the original facial recognition model.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence machine learning technology, and in particular to a facial feature updating method, apparatus, computer equipment and storage medium. Background Art

[0002] With the widespread use of facial recognition algorithms in various scenarios, recognition systems based on facial recognition algorithms are being applied in more and more scenarios. Although the accuracy of the current recognition system has reached a certain level, in application scenarios with a higher security level, even a relatively low error rate can have a huge impact on actual production. For example, although the overall facial recognition error rate is already at a low level, the facial features of a few individuals are easily confused with those of other individuals, resulting in the recognition error rate of this group being much higher than the overall error rate, which in turn affects the accuracy of the recognition system.

[0003] Therefore, how to further reduce the misrecognition rate while ensuring the overall accuracy is a problem that needs to be solved urgently. Summary of the invention

[0004] Based on this, it is necessary to provide a facial feature updating method, device, computer equipment and storage medium that can improve the accuracy in order to solve the above technical problems.

[0005] A facial feature updating method, the method comprising:

[0006] Determine a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the objects to which the facial images in the facial update set belong do not overlap with the objects to which the facial images in the facial image base library belong;

[0007] Training a facial recognition model based on the facial image test library to obtain an error recognition rate;

[0008] When the false recognition rate meets the facial feature update condition, facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set are extracted respectively, and feature update is performed on the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features.

[0009] In one embodiment, the updating of features of corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features includes:

[0010] Based on the extracted facial features, determining a first inter-class similarity, a second inter-class similarity, and an intra-class similarity;

[0011] Determine an inter-class difference between at least one of the first inter-class similarity and the second inter-class similarity and an inter-class threshold, and an intra-class difference between the intra-class similarity and the intra-class threshold;

[0012] With the goal of minimizing the false recognition rate, the facial features corresponding to the intra-class similarity are updated until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, thereby obtaining the updated facial features;

[0013] Based on the updated facial features, the corresponding facial images in the facial image base library and the facial image test library are updated.

[0014] In one of the embodiments, the facial feature update condition includes that the overall error recognition rate is less than the error recognition rate threshold, and the individual error recognition rate is greater than the overall error recognition rate.

[0015] In one embodiment, extracting facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set, and determining the first inter-class similarity, the second inter-class similarity, and the intra-class similarity based on the extracted facial features, comprises:

[0016] Extract facial features from the facial base database, the facial update set, and the facial test database respectively to obtain facial base database features, facial update set features, and facial test database features;

[0017] According to the similarity between the facial base database features and the facial update set features, the first inter-class similarity is obtained;

[0018] According to the similarity between the features of the face update set and the features of the face test library, the second inter-class similarity is obtained;

[0019] According to the similarity between the features of the facial base library and the features of the facial test library, the intra-class similarity is obtained.

[0020] In one embodiment, the training of the facial recognition model based on the facial image test library to obtain the error recognition rate includes:

[0021] Extracting facial feature vectors of each facial image in the facial image test library;

[0022] Transmitting the extracted facial feature vector to a server, so that the server recognizes the facial feature vector using a facial recognition model to obtain a facial recognition prediction result;

[0023] The triggering server compares the facial recognition prediction result with the corresponding facial recognition actual result to obtain a comparison result, and obtains an error recognition rate based on the comparison result; and receives the error recognition rate returned by the server.

[0024] In one embodiment, extracting the facial feature vector of each facial image in the facial image test library includes:

[0025] Detect key facial feature points in facial images and locate key facial feature points;

[0026] A deep convolutional neural network is used to extract key facial feature points and obtain facial feature vectors.

[0027] In one embodiment, the facial recognition model training step includes:

[0028] The server obtains the facial images in the facial image test library, and performs facial detection processing on the facial images to obtain facial images to be identified;

[0029] The server performs facial alignment processing on the facial image to be recognized to obtain an aligned facial image, and inputs the aligned facial image into a facial recognition model, and the facial recognition model outputs a facial recognition prediction result;

[0030] The server constructs a loss function based on the difference between the facial recognition prediction result and the corresponding facial recognition actual result, and trains the facial recognition model with the goal of minimizing the loss function until the training termination condition is met, and a trained facial recognition model is obtained.

[0031] In one embodiment, the server uses a facial recognition model to identify the facial feature vector to obtain a facial recognition prediction result, including:

[0032] The server calculates the similarity between the facial feature vector and the corresponding feature vector in the facial image base library;

[0033] The server outputs a facial recognition prediction result based on the comparison result between the similarity and the similarity threshold.

[0034] A facial feature updating device, comprising:

[0035] A determination module is used to determine a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the object to which the facial image in the facial update set belongs does not overlap with the object to which the facial image in the facial image base library belongs;

[0036] A training module, used for training a facial recognition model based on the facial image test library to obtain an error recognition rate;

[0037] The updating module is used for respectively extracting facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set when the false recognition rate meets the facial feature updating condition, and performing feature update on the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features.

[0038] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the following method are implemented:

[0039] Determine a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the objects to which the facial images in the facial update set belong do not overlap with the objects to which the facial images in the facial image base library belong;

[0040] Training a facial recognition model based on the facial image test library to obtain an error recognition rate;

[0041] When the false recognition rate meets the facial feature update condition, facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set are extracted respectively, and feature update is performed on the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features.

[0042] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following method steps:

[0043] Determine a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the objects to which the facial images in the facial update set belong do not overlap with the objects to which the facial images in the facial image base library belong;

[0044] Training a facial recognition model based on the facial image test library to obtain an error recognition rate;

[0045] When the false recognition rate meets the facial feature update condition, facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set are extracted respectively, and feature update is performed on the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features.

[0046] The above-mentioned facial feature updating method, device, computer equipment and storage medium first determine the facial image base library, facial image test library and facial image update set, and by constructing an external unlabeled facial data set, the original image acquisition and detection function module or hardware module can be used to update the features, without the need for manual acquisition and labeling, thereby greatly reducing the labor cost and time cycle; then, by obtaining the error recognition rate indicator during the facial recognition model training process, and when the error recognition rate meets the facial feature update condition, by extracting the image features of the facial image base library, facial image test library and facial image update set, the facial image base library and facial image test library are updated, and the facial recognition algorithm can be upgraded during the operation of the facial recognition system, and the facial recognition system can be updated and iterated quickly and conveniently without retraining the original facial recognition model; at the same time, the facial recognition model is updated by using the unlabeled facial image data set, so that the variance of the false alarm rate of different individuals can be kept stable under the premise of ensuring the accuracy of the original facial recognition model, and the false alarm rate of facial recognition can be further quickly reduced, thereby improving the accuracy of facial recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1A A diagram showing an application environment of a facial feature updating method in an embodiment;

[0048] Figure 1B A diagram showing an application environment of a facial feature updating method according to another embodiment;

[0049] Figure 2 is a flowchart of a facial feature updating method in one embodiment;

[0050] Figure 3 is a flowchart of steps for a computer device to perform feature update based on extracted facial features in one embodiment;

[0051] Figure 4 A schematic diagram of the principle of updating features of a computer device in one embodiment;

[0052] Figure 5 is a flowchart of steps for a computer device to perform similarity calculation based on extracted facial features in one embodiment;

[0053] Figure 6 is a flowchart of steps for a computer device to train a facial recognition model based on the facial image test library in an embodiment;

[0054] Figure 7 A schematic diagram of a flow chart of steps for training a facial recognition model at a server in one embodiment;

[0055] Figure 8A schematic diagram of the principle of online feature updating of a facial recognition system in one embodiment;

[0056] Fig. 9 is a structural block diagram of a facial feature updating device in one embodiment;

[0057] Fig.10 1 is a diagram of the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0059] With the development of deep convolutional neural networks and large-scale data sets, deep facial recognition has made significant progress and has been widely used in real-world applications. Facial recognition is a biometric technology that uses facial feature information to identify people. Usually, a camera or a camera is used to capture images or video streams containing faces, and automatically detects and tracks faces in the images, and then recognizes the detected faces. The entire facial recognition system usually consists of four key elements: facial image acquisition and detection, facial image preprocessing, facial image feature extraction, and matching and recognition. These four elements are all implemented by deep convolutional neural networks. Because facial algorithms are affected by many factors such as backlight, dark light, strong light, and recognition angle when used in products, how to quickly iterate algorithm products, improve performance, and quickly deploy them has become an important link in practical applications.

[0060] Currently, the following methods are usually used to optimize the performance of facial recognition systems: first, through terminal hardware optimization, mainly focusing on the foreground module of the camera from the image acquisition to solve problems such as backlighting or clarity; second, retraining the facial recognition algorithm model to optimize the accuracy of the algorithm itself. However, the former requires hardware improvements, which is costly and has poor applicability to different products. The latter requires retraining, which takes a long time, has high labor costs, and makes product iteration difficult.

[0061] In view of this, the present application provides a facial feature updating method, apparatus, computer device and storage medium, which can upgrade the facial algorithm and improve the accuracy without changing the deep convolutional neural network or retraining the network model.

[0062] The facial feature updating method provided in this application can be applied to Figure 1AIn the application environment shown. Among them, the camera device 110 is connected to the computer device 120 through a network. The camera device 110 is used to capture facial images and send the facial images to the computer device 120; the computer device receives the facial images transmitted by the camera device 110 and recognizes the facial images. Among them, the camera device 110 includes but is not limited to a surveillance camera, or a camera module integrated in an electronic device, etc. The computer device 120 can be a terminal or a server, and the server can be implemented with an independent server or a server cluster composed of multiple servers. Terminals include but are not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices.

[0063] In some embodiments, the facial feature updating method provided by the present application can also be applied to Figure 1B In the application environment shown in the figure, the computer device 120 is also connected to the server 130 through a network. The computer device receives the facial image transmitted by the camera device 110 and extracts the facial features in the facial image; the computer device 120 transmits the facial features to the server 130, and the server 130 recognizes and returns the recognition result to the computer device 120. The server 130 can be understood as a remote server or a cloud server.

[0064] In the above embodiment, the most computationally intensive phase of facial recognition is completed on the computer device 120, reducing the performance requirements on the server 130. At the same time, the facial image base library and the facial image test library can be stored on the server 130, avoiding the risk of facial image leakage on the computer device 120 and protecting the privacy of facial images. At the same time, the computer device 120 only transmits facial features to the server 130, reducing the amount of transmitted data and improving transmission efficiency.

[0065] In one embodiment, Figure 2 As shown, a facial feature updating method is provided, which is described by taking the method applied to the computer device 120 in FIG. 1 as an example, and includes the following steps:

[0066] Step S202, determining a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the object to which the facial image in the facial update set belongs does not overlap with the object to which the facial image in the facial image base library belongs.

[0067] The terms "facial image base library" and "facial image test library" refer to a pre-established data set containing multiple facial images with labeled object identifiers, wherein the facial image base library is usually used for 1:N comparison of facial features, and the facial image test library is usually used for facial 1:1 test or 1:N test. The images in the facial image base library and the facial image test library have a mapping relationship, that is, the two libraries include facial images belonging to the same object.

[0068] The term "object" refers to an individual for which facial recognition is performed, including a person, an animal, etc. Object identification is used to distinguish different objects, including but not limited to object ID, etc.

[0069] The term "face image update set" refers to a data set containing multiple face images without object identification, which is an image data set used to train a neural network for feature update. In some embodiments, during operation, the camera hardware randomly performs face detection on the detected images. If it is determined that the object to which the detected face image belongs does not overlap with the object to which the face image in the face image base library belongs (in other words, the object to which the detected face image belongs has not appeared in the face image base library), the detected face image is placed in the face image update set, thereby constructing a face update set.

[0070] Specifically, the computer device obtains the facial image base library, the facial image test library, and the facial image update set from the database for subsequent training steps. In some embodiments, the computer device can also download from the cloud to obtain the latest image data.

[0071] Step S204, training the facial recognition model based on the facial image test library to obtain the false recognition rate.

[0072] The false recognition rate is the ratio of the number of times that the matching should have failed but was mistakenly identified as a successful matching to the total number of inter-class matchings.

[0073] Specifically, the computer device trains the facial recognition model based on the facial image test library and obtains the error recognition rate during the training process. For example, the computer device inputs part or all of the images in the facial image test library into the facial recognition model, and compares them with the facial images in the facial image base library, and the facial recognition model outputs the facial recognition prediction result. The computer device calculates the error recognition rate based on the difference and number between the facial recognition prediction result and the actual result.

[0074] In some embodiments, the server can also complete the recognition and comparison steps to reduce the pressure on the computer device. For example, the computer device extracts the facial features of each facial image in the facial image test library and transmits the facial features to the server, which completes the training and returns the result of the error recognition rate.

[0075] Step S206, when the false recognition rate meets the facial feature update condition, facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set are extracted respectively, and the features of the corresponding facial images in the facial image base library and the facial image test library are updated based on the extracted facial features.

[0076] Among them, the facial feature update conditions include that the overall error recognition rate is less than the error recognition rate threshold, and the individual error recognition rate is greater than the overall error recognition rate.

[0077] Specifically, the computer device determines whether the error recognition rate meets the facial feature update condition based on the error recognition rate. If it is determined that the error recognition rate meets the facial feature update condition, the computer device extracts facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set respectively. Among them, the computer device can use a deep convolutional neural network to perform the step of extracting facial features. According to the facial base library features extracted from each facial image in the facial image base library, the facial test library features extracted from each facial image in the facial image test library, and the facial update set features extracted from each facial image in the facial image update set, the computer device uses these facial features for training to obtain updated facial features, and updates the corresponding facial images in the facial image base library and the facial image test library with the updated facial features, thereby updating the features of the corresponding facial images in the facial image base library and the facial image test library.

[0078] The above-mentioned facial feature updating method obtains the false recognition rate index during the facial recognition model training process, and when the false recognition rate meets the facial feature updating condition, based on the pre-established facial image base library and facial image test library, and the facial image update set, extracts the intra-class similarity between the facial features of facial images belonging to the same object, and the inter-class similarity between the facial features of facial images belonging to different objects, and performs training with the goal of minimizing the false recognition rate, and after updating the facial features, updates the facial image base library and the facial image test library. Thus, under the premise of ensuring the accuracy of the original facial recognition model, the false alarm rate can be quickly further reduced without retraining the original facial recognition model, thereby improving the accuracy rate.

[0079] In some embodiments, Figure 3 As shown, the step of updating the features of the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features by the computer device includes:

[0080] Step S302: determining the first inter-class similarity, the second inter-class similarity, and the intra-class similarity based on the extracted facial features.

[0081] Specifically, the computer device calculates the inter-class similarity and the intra-class similarity based on the facial base library features extracted from each facial image in the facial image base library, the facial test library features extracted from each facial image in the facial image test library, and the facial update set features extracted from each facial image in the facial image update set, and obtains the intra-class similarity between the facial base library features and the facial test library features, the first inter-class similarity between the facial base library features and the facial update set features, and the second inter-class similarity between the facial test library features and the facial update set features, so as to subsequently perform feature update according to the intra-class similarity and the inter-class similarity.

[0082] Step S304: determining an inter-class difference between at least one of the first inter-class similarity and the second inter-class similarity and an inter-class threshold, and an intra-class difference between the intra-class similarity and the intra-class threshold.

[0083] Specifically, the computer device calculates the difference between the first inter-class similarity and the inter-class threshold as the inter-class difference based on the pre-set inter-class threshold and intra-class threshold; and calculates the difference between the intra-class similarity and the intra-class threshold as the intra-class difference. Alternatively, the computer device calculates the inter-class difference between the second inter-class similarity and the inter-class threshold. Alternatively, the computer device calculates the inter-class difference between the first inter-class similarity and the inter-class threshold and the inter-class difference between the second inter-class similarity and the inter-class threshold, and combines the two inter-class differences to obtain the final inter-class difference. For example, the mean of the two inter-class differences is used as the final inter-class difference, and for example, the smaller value of the two inter-class differences is used as the final inter-class difference, and so on.

[0084] Step S306, with the goal of minimizing the false recognition rate, the facial features corresponding to the intra-class similarity are updated until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, to obtain the updated facial features.

[0085] Among them, for the facial image base library and the facial image test library, since the facial images therein belong to the same object, the smaller the intra-class similarity, the more accurate it is. For the facial image base library and the facial image update set, as well as the facial image test library and the facial image update set, since the facial images therein belong to different objects, the larger the inter-class similarity, the more accurate it is. Therefore, specifically, after the computer device calculates the intra-class difference and the inter-class difference, the facial features corresponding to the intra-class similarity are updated with the goal of minimizing the false recognition rate, that is, the facial features of the facial images belonging to the same object in the facial image base library and the facial image test library are updated until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, and the updated facial features are obtained.

[0086] In a specific example, Figure 4As shown, the computer device uses a backbone network (such as a deep convolutional neural network) to extract features from the facial image base library A, the facial image update set B, and the facial image test library C, respectively, to obtain facial base library features A, facial update set features B, and facial test library features C, and obtains the inter-class similarity (A, B) between the facial base library features A and the facial update set features B (i.e., the first inter-class similarity), the intra-class similarity (A, C) between the facial base library features A and the facial test library features C, and the inter-class similarity (B, C) between the facial test library features C and the facial update set features B (i.e., the first inter-class similarity). According to the inter-class similarity and the intra-class similarity, the facial features corresponding to the intra-class similarity are updated to obtain the updated facial features.

[0087] In some embodiments, in order to avoid the situation where the overall error rate meets the error rate threshold, but the error rate of a few individuals is much higher than the overall error rate, thereby affecting the accuracy of the recognition system, the computer device can also update the facial features until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the overall error rate is less than the previous overall error rate.

[0088] Step S308: Based on the updated facial features, the corresponding facial images in the facial image base library and the facial image test library are updated.

[0089] Specifically, based on the updated facial features, the computer device replaces the corresponding facial images in the facial image base library and the facial image test library with the facial images corresponding to the updated facial features, thereby completing the update of the facial image base library and the facial image test library.

[0090] In some embodiments, Figure 5 As shown, the steps of extracting facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set by the computer device respectively, and determining the first inter-class similarity, the second inter-class similarity, and the intra-class similarity based on the extracted facial features include:

[0091] Step S502, extracting facial features from the facial base database, facial update set, and facial test database respectively, to obtain facial base database features, facial update set features, and facial test database features.

[0092] Step S504, obtaining a first inter-class similarity based on the similarity between the facial base database features and the facial update set features.

[0093] Step S506, obtaining a second inter-class similarity based on the similarity between the face update set features and the face test library features.

[0094] Step S508, obtaining intra-class similarity based on the similarity between the facial base library features and the facial test library features.

[0095] Specifically, the computer device extracts facial features from the facial base database, the facial update set, and the facial test database, respectively, to obtain facial base database features, facial update set features, and facial test database features. For facial base database features and facial test database features belonging to the same object, the similarity between the two features is calculated to obtain intra-class similarity. The similarity between the two features can usually be calculated by calculating the distance between the feature vectors of the two features. The distance between the feature vectors includes Euclidean distance, cosine distance, or trigonometric distance, etc.

[0096] Since the objects to which the facial images in the face update set belong are different from the objects to which the facial images in the face base library and the face test library belong, the computer device calculates the similarity between the features of the face base library / face test library and the features of the face update set, and obtains the inter-class similarity. Specifically, the computer device obtains the first inter-class similarity based on the similarity between the features of the face base library and the features of the face update set; and obtains the second inter-class similarity based on the similarity between the features of the face update set and the features of the face test library.

[0097] In the above embodiment, based on the similarity between each pair of features in the facial base library features, facial test library features and facial update set features, the inter-class similarity and intra-class similarity are determined, and feature update can be performed accordingly, making facial recognition more accurate.

[0098] In some embodiments, Figure 6 As shown, the computer device trains the facial recognition model based on the facial image test library to obtain the error recognition rate, including:

[0099] Step S602, extracting facial feature vectors of each facial image in the facial image test library.

[0100] Step S604, transmitting the extracted facial feature vector to the server, so that the server recognizes the facial feature vector using a facial recognition model to obtain a facial recognition prediction result.

[0101] Step S606, triggering the server to compare the facial recognition prediction result with the corresponding facial recognition actual result to obtain a comparison result, and obtain an error recognition rate based on the comparison result.

[0102] Step S608: receiving the error recognition rate returned by the server.

[0103] Specifically, the computer device extracts the facial feature vector of each facial image in the facial image test library and transmits the extracted facial feature vector to the server. For example, the computer device extracts the facial feature vector of each facial image in the facial image test library using a deep convolutional neural network.

[0104] In some embodiments, the step of extracting facial feature vectors of each facial image in the facial image test library by a computer device includes: detecting key facial feature points in the facial image and locating the key facial feature points; extracting the key facial feature points using a deep convolutional neural network to obtain a facial feature vector. Specifically, during the feature point detection process, the computer device automatically locates the key facial feature points based on the input facial image. For example, if the object is a person, the key facial feature points include eyes, nose tip, mouth corners, eyebrows, and other component contour points. Then, the computer device extracts the key facial feature points using a deep convolutional neural network to obtain a facial feature vector. For example, the computer device extracts the facial feature vector using a feature extractor and transmits the extracted facial feature vector to the server.

[0105] After receiving the facial feature vector, the server uses the facial recognition model to identify the facial feature vector and obtain the facial recognition prediction result; the server then compares the facial recognition prediction result with the corresponding facial recognition actual result to obtain the comparison result, and obtains the error recognition rate based on the comparison result. The server then sends the error recognition rate to the computer device, and the computer device receives the error recognition rate returned by the server.

[0106] In some embodiments, the server uses a facial recognition model to identify the facial feature vector to obtain a facial recognition prediction result, including: the server calculates the similarity between the facial feature vector and the corresponding feature vector in the facial image base library; outputs the facial recognition prediction result according to the comparison result of the similarity and the similarity threshold. Exemplarily, the server can compare the facial feature vector with the facial image base library based on the facial feature vector, which may include: calculating the Euclidean distance between the facial feature vector and the feature vector of the corresponding facial image in the facial image base library. If the Euclidean distance between one or more feature vectors in the facial image base library and the facial feature vector is less than or equal to the threshold, the one or more feature vectors can be output as the prediction result. Furthermore, the feature vector with the highest matching degree (i.e., the smallest Euclidean distance with the extracted facial feature vector) among the one or more feature vectors can be output as the prediction result.

[0107] In the above embodiment, the recognition and comparison steps are completed by the server, and the most computationally intensive phase of facial recognition is completed on the computer device, which reduces the performance requirements on the server. At the same time, the facial image base library and the facial image test library can be stored on the server, avoiding the risk of facial image leakage on the computer device and protecting the privacy and security of facial images. At the same time, the computer device only transmits facial features to the server, reducing the amount of transmitted data and improving transmission efficiency.

[0108] In order to reduce the hardware requirements of the front-end device, the training process of the facial recognition model can be performed by the server. Figure 7 As shown, the training steps of the facial recognition model include:

[0109] Step S702: the server obtains facial images from a facial image test library, and performs facial detection processing on the facial images to obtain facial images to be identified.

[0110] Specifically, the server obtains a facial image in a facial image test library and performs facial detection on the facial image. When a face is detected in the facial image, the facial image is obtained as a facial image to be identified.

[0111] Step S704: the server performs facial alignment processing on the facial image to be recognized, obtains the facial image after the alignment processing, and inputs the facial image after the alignment processing into the facial recognition model, and the facial recognition model outputs the facial recognition prediction result.

[0112] Among them, the main purpose of alignment processing is to allow the computer to automatically locate the positions of various parts of the face (such as the corners of the eyes, the corners of the mouth, the tip of the nose, etc.). In order to reduce the difficulty of recognition, the face of the subject is rotated to the same angle through alignment processing to improve the accuracy of facial recognition.

[0113] Specifically, the server performs facial alignment processing on the facial image to be recognized to obtain the facial image after the alignment processing. Then, the server inputs the facial image after the alignment processing into the facial recognition model, and the facial recognition model outputs the facial recognition prediction result.

[0114] In step S706, the server constructs a loss function based on the difference between the facial recognition prediction result and the corresponding facial recognition actual result, and trains the facial recognition model with the goal of minimizing the loss function until the training termination condition is met, and a trained facial recognition model is obtained.

[0115] Specifically, the server constructs a loss function based on the difference between the facial recognition prediction result and the corresponding facial recognition actual result (for example, taking the corresponding facial image in the facial image base library as the facial recognition actual result, or taking the pre-annotated label carried by the image in the facial image test library as the facial recognition actual result, etc.), and trains the facial recognition model with the goal of minimizing the loss function until the training termination condition is met, and a trained facial recognition model is obtained. Among them, the training termination condition includes but is not limited to the number of training times reaching a preset threshold, the training time reaching a preset threshold, and the error rate being less than the error rate threshold.

[0116] In the above embodiment, by deploying the facial recognition model on the server side, the computing pressure of the computer device can be reduced; at the same time, the facial recognition model is trained by the server side based on facial images, and the facial image base library and the facial image test library can be stored on the server side, thereby avoiding the risk of facial image leakage on the computer device and protecting the privacy and security of facial images.

[0117] In a specific example, Figure 8 As shown, the embodiments of the present application can also be applied to facial recognition systems. After the camera device acquires the image, the computer device performs image detection and preprocessing, and performs feature extraction, and sends the extracted features to the back-end server (i.e., the server in the aforementioned embodiment) for the server to perform 1:1 comparison or 1:N comparison. When the computer device detects that the individual misrecognition rate of the facial recognition system is high, the online feature update module is turned on, and the online feature update is performed based on the feature update library (i.e., the facial feature update set in the aforementioned embodiment) constructed by the camera device through random image extraction, without the need to retrain the facial recognition model of the facial recognition system.

[0118] In the above embodiment, when the facial recognition system is working, it can solve the problem of too high individual false alarm rate through the software switch feature update module, and easily update and iterate the system without retraining the artificial intelligence algorithm model. At the same time, the original hardware modules of image acquisition and image detection are used to build an external unlabeled facial data set, without the need for manual acquisition and labeling, which greatly reduces labor costs and time cycles. In the process of feature update, by adjusting the facial feature density distribution to ensure uniform feature density distribution, feature update alignment is performed based on the unlabeled facial data set, so that the facial recognition algorithm has higher accuracy and lower false alarm rate. At the same time, the variance of evaluation indicators such as false alarm rate of different individuals remains stable and will not produce large differences.

[0119] It should be understood that although Figure 2-3The steps in the flowcharts of 5-7 are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2-3 , at least part of the steps in 5-7 may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0120] In one embodiment, Fig. 9 As shown, a facial feature updating device 900 is provided, comprising: a determination module 901, a training module 902, and an updating module 903, wherein:

[0121] The determination module 901 is used to determine a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the objects to which the facial images in the facial update set belong do not overlap with the objects to which the facial images in the facial image base library belong.

[0122] The training module 902 is used to train the facial recognition model based on the facial image test library to obtain the false recognition rate.

[0123] The updating module 903 is used to extract the facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set respectively when the false recognition rate meets the facial feature update condition, and update the features of the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features.

[0124] In one embodiment, the update module 903 is also used to determine a first inter-class similarity, a second inter-class similarity, and an intra-class similarity based on the extracted facial features; determine the inter-class difference between at least one of the first inter-class similarity and the second inter-class similarity and an inter-class threshold, and the intra-class difference between the intra-class similarity and the intra-class threshold; with the goal of minimizing the false recognition rate, update the facial features corresponding to the intra-class similarity until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, to obtain updated facial features; and update the corresponding facial images in the facial image base library and the facial image test library based on the updated facial features.

[0125] In one embodiment, the facial feature update condition includes that the overall error recognition rate is less than the error recognition rate threshold, and the individual error recognition rate is greater than the overall error recognition rate.

[0126] In one embodiment, the update module 903 is further used to extract facial features from a facial base library, a facial update set, and a facial test library, respectively, to obtain facial base library features, facial update set features, and facial test library features; obtain a first inter-class similarity based on the similarity between the facial base library features and the facial update set features; obtain a second inter-class similarity based on the similarity between the facial update set features and the facial test library features; and obtain an intra-class similarity based on the similarity between the facial base library features and the facial test library features.

[0127] In one embodiment, the training module 902 is also used to train the facial recognition model based on the facial image test library to obtain the error recognition rate, including: extracting the facial feature vector of each facial image in the facial image test library; transmitting the extracted facial feature vector to the server, so that the server uses the facial recognition model to identify the facial feature vector to obtain a facial recognition prediction result; triggering the server to compare the facial recognition prediction result with the corresponding facial recognition actual result to obtain a comparison result, and obtain the error recognition rate based on the comparison result; and receiving the error recognition rate returned by the server.

[0128] In one embodiment, the training module 902 is also used to detect key facial feature points in the facial image and locate the key facial feature points; and to extract the key facial feature points using a deep convolutional neural network to obtain a facial feature vector.

[0129] In one of the embodiments, the server is used to obtain facial images in a facial image test library, and perform facial detection processing on the facial images to obtain facial images to be identified; perform facial alignment processing on the facial images to be identified to obtain facial images after alignment processing, and input the facial images after alignment processing into a facial recognition model, and the facial recognition model outputs facial recognition prediction results; based on the difference between the facial recognition prediction results and the corresponding facial recognition actual results, a loss function is constructed, and the facial recognition model is trained with the goal of minimizing the loss function until the training is terminated when the training termination condition is met, and a trained facial recognition model is obtained.

[0130] In one of the embodiments, the server is also used to calculate the similarity between the facial feature vector and the corresponding feature vector in the facial image base library; and output the facial recognition prediction result based on the comparison result of the similarity and the similarity threshold.

[0131] For the specific definition of the facial feature updating device, please refer to the definition of the facial feature updating method above, which will not be repeated here. Each module in the above facial feature updating device can be implemented in whole or in part by software, hardware and their combination. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0132] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store facial image data, etc. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a facial feature updating method is implemented.

[0133] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0134] In one embodiment, a computer device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the following steps when executing the computer program: determining a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the object to which the facial image in the facial update set belongs does not overlap with the object to which the facial image in the facial image base library belongs; training a facial recognition model based on the facial image test library to obtain an error recognition rate; when the error recognition rate meets a facial feature update condition, extracting facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set, respectively, and performing feature update on the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features.

[0135] In one embodiment, the processor further implements the following steps when executing the computer program: based on the extracted facial features, determine the first inter-class similarity, the second inter-class similarity, and the intra-class similarity; determine the inter-class difference between at least one of the first inter-class similarity and the second inter-class similarity and the inter-class threshold, and the intra-class difference between the intra-class similarity and the intra-class threshold; with the goal of minimizing the false recognition rate, update the facial features corresponding to the intra-class similarity until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, to obtain the updated facial features; based on the updated facial features, update the corresponding facial images in the facial image base library and the facial image test library.

[0136] In one embodiment, when the processor executes the computer program, the following steps are also implemented: facial features in the face base library, the face update set, and the face test library are extracted respectively to obtain facial base library features, facial update set features, and facial test library features; a first inter-class similarity is obtained based on the similarity between the facial base library features and the facial update set features; a second inter-class similarity is obtained based on the similarity between the facial update set features and the facial test library features; and an intra-class similarity is obtained based on the similarity between the facial base library features and the facial test library features.

[0137] In one embodiment, the processor further implements the following steps when executing the computer program: extracting facial feature vectors of each facial image in a facial image test library; transmitting the extracted facial feature vectors to a server, so that the server recognizes the facial feature vectors using a facial recognition model to obtain a facial recognition prediction result; triggering the server to compare the facial recognition prediction result with the corresponding facial recognition actual result to obtain a comparison result, and obtaining an error recognition rate based on the comparison result; and receiving the error recognition rate returned by the server.

[0138] In one embodiment, when the processor executes the computer program, the following steps are also implemented: detecting key facial feature points in the facial image and locating the key facial feature points; and extracting the key facial feature points using a deep convolutional neural network to obtain a facial feature vector.

[0139] The computer device obtains the error recognition rate index during the training process of the facial recognition model, and when the error recognition rate meets the facial feature update condition, based on the pre-established facial image base library, facial image test library, and facial image update set, extracts the intra-class similarity between facial features of facial images belonging to the same object, and the inter-class similarity between facial features of facial images belonging to different objects, and performs training with the goal of minimizing the error recognition rate, and updates the facial image base library and facial image test library after updating the facial features. Thus, under the premise of ensuring the accuracy of the original facial recognition model, the false alarm rate can be quickly further reduced without retraining the original facial recognition model, thereby improving the accuracy rate.

[0140] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: determining a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the object to which the facial image in the facial update set belongs does not overlap with the object to which the facial image in the facial image base library belongs; training a facial recognition model based on the facial image test library to obtain an error recognition rate; when the error recognition rate meets a facial feature update condition, extracting facial features of each facial image in the facial image base library, the facial image test library, and the facial image update set, respectively, and updating the features of the corresponding facial images in the facial image base library and the facial image test library based on the extracted facial features.

[0141] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: based on the extracted facial features, a first inter-class similarity, a second inter-class similarity, and an intra-class similarity are determined; an inter-class difference between at least one of the first inter-class similarity and the second inter-class similarity and an inter-class threshold, and an intra-class difference between the intra-class similarity and the intra-class threshold are determined; with the goal of minimizing the false recognition rate, the facial features corresponding to the intra-class similarity are updated until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, to obtain updated facial features; based on the updated facial features, the corresponding facial images in the facial image base library and the facial image test library are updated.

[0142] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: facial features are extracted from the face base library, the face update set, and the face test library respectively to obtain face base library features, face update set features, and face test library features; a first inter-class similarity is obtained based on the similarity between the face base library features and the face update set features; a second inter-class similarity is obtained based on the similarity between the face update set features and the face test library features; and an intra-class similarity is obtained based on the similarity between the face base library features and the face test library features.

[0143] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: extracting facial feature vectors of each facial image in the facial image test library; transmitting the extracted facial feature vectors to the server, so that the server uses a facial recognition model to identify the facial feature vectors to obtain facial recognition prediction results; triggering the server to compare the facial recognition prediction results with the corresponding facial recognition actual results to obtain comparison results, and obtaining an error recognition rate based on the comparison results; and receiving the error recognition rate returned by the server.

[0144] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: detecting key facial feature points in the facial image and locating the key facial feature points; and extracting the key facial feature points using a deep convolutional neural network to obtain a facial feature vector.

[0145] The computer-readable storage medium obtains the error recognition rate index during the training process of the facial recognition model, and when the error recognition rate meets the facial feature update condition, based on the pre-established facial image base library and facial image test library, and the facial image update set, extracts the intra-class similarity between the facial features of facial images belonging to the same object, and the inter-class similarity between the facial features of facial images belonging to different objects, and performs training with the goal of minimizing the error recognition rate, and updates the facial image base library and the facial image test library after updating the facial features. Thus, under the premise of ensuring the accuracy of the original facial recognition model, the false alarm rate can be quickly further reduced without retraining the original facial recognition model, thereby improving the accuracy rate.

[0146] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0147] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0148] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A facial feature updating method, characterized in that: The method comprises: Determine a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the objects to which the facial images in the facial image update set belong do not overlap with the objects to which the facial images in the facial image base library belong; Training a facial recognition model based on the facial image test library to obtain an error recognition rate; When the error recognition rate satisfies the facial feature update condition, facial features in the facial image base library, the facial image update set, and the facial image test library are extracted respectively to obtain facial base library features, facial update set features, and facial test library features; the facial feature update condition includes that the overall error recognition rate is less than the error recognition rate threshold, and the individual error recognition rate is greater than the overall error recognition rate; Obtaining a first inter-class similarity according to the similarity between the facial base database features and the facial update set features; Obtaining a second inter-class similarity according to the similarity between the facial update set features and the facial test library features; Obtaining intra-class similarity according to the similarity between the facial base library features and the facial test library features; Determine an inter-class difference between at least one of the first inter-class similarity and the second inter-class similarity and an inter-class threshold, and an intra-class difference between the intra-class similarity and an intra-class threshold; With the goal of minimizing the false recognition rate, the facial features corresponding to the intra-class similarity are updated until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, thereby obtaining the updated facial features; Based on the updated facial features, corresponding facial images in the facial image base library and the facial image test library are updated.

2. The method according to claim 1, characterized in that The facial recognition model is trained based on the facial image test library to obtain the false recognition rate, including: Extracting facial feature vectors of each facial image in the facial image test library; Transmitting the extracted facial feature vector to a server, so that the server recognizes the facial feature vector using a facial recognition model to obtain a facial recognition prediction result; The triggering server compares the facial recognition prediction result with the corresponding facial recognition actual result to obtain a comparison result, and obtains an error recognition rate based on the comparison result; and receives the error recognition rate returned by the server.

3. The method according to claim 2, characterized in that The step of extracting a facial feature vector of each facial image in the facial image test library comprises: Detect key facial feature points in facial images and locate key facial feature points; A deep convolutional neural network is used to extract key facial feature points and obtain facial feature vectors.

4. The method according to claim 2, characterized in that: The training steps of the facial recognition model include: The server obtains the facial images in the facial image test library, and performs facial detection processing on the facial images to obtain facial images to be identified; The server performs facial alignment processing on the facial image to be recognized to obtain an aligned facial image, and inputs the aligned facial image into a facial recognition model, and the facial recognition model outputs a facial recognition prediction result; The server constructs a loss function based on the difference between the facial recognition prediction result and the corresponding facial recognition actual result, and trains the facial recognition model with the goal of minimizing the loss function until the training termination condition is met, and a trained facial recognition model is obtained.

5. The method according to claim 4, characterized in that The method further comprises: The corresponding facial images in the facial image base library are used as actual facial recognition results, or the pre-labeled labels carried by the images in the facial image test library are used as actual facial recognition results.

6. The method according to claim 2, characterized in that The server uses a facial recognition model to identify the facial feature vector to obtain a facial recognition prediction result, including: The server calculates the similarity between the facial feature vector and the corresponding feature vector in the facial image base library; The server outputs a facial recognition prediction result based on the comparison result between the similarity and the similarity threshold.

7. The method according to claim 6, characterized in that The similarity is characterized by Euclidean distance; and outputting a facial recognition prediction result according to a comparison result between the similarity and a similarity threshold value includes: If the Euclidean distance between one or more feature vectors in the facial image base and the facial feature vector is less than or equal to a threshold, the one or more feature vectors are output as facial recognition prediction results.

8. The method according to claim 1, characterized in that: The updating of the corresponding facial images in the facial image base library and the facial image test library based on the updated facial features comprises: The facial images corresponding to the updated facial features are used to replace the corresponding facial images in the facial image base library and the facial image test library, so as to complete the updating of the facial image base library and the facial image test library.

9. A facial feature updating device, characterized in that: The device comprises: A determination module is used to determine a facial image base library, a facial image test library, and a facial image update set; wherein the facial image base library and the facial image test library include facial images belonging to the same object, and the objects to which the facial images in the facial image update set belong do not overlap with the objects to which the facial images in the facial image base library belong; A training module, used for training a facial recognition model based on the facial image test library to obtain an error recognition rate; An updating module is used for extracting facial features from a facial image base library, a facial image update set, and a facial image test library respectively when the false recognition rate meets a facial feature update condition, and obtaining facial base library features, facial update set features, and facial test library features; the facial feature update condition includes that the overall false recognition rate is less than a false recognition rate threshold, and the individual false recognition rate is greater than the overall false recognition rate; obtaining a first inter-class similarity based on the similarity between the facial base library features and the facial update set features; obtaining a second inter-class similarity based on the similarity between the facial update set features and the facial test library features; and obtaining an intra-class similarity based on the similarity between the facial base library features and the facial test library features; Determine the inter-class difference between at least one of the first inter-class similarity and the second inter-class similarity and the inter-class threshold, and the intra-class difference between the intra-class similarity and the intra-class threshold; with the goal of minimizing the false recognition rate, update the facial features corresponding to the intra-class similarity until the inter-class difference is greater than the inter-class threshold, the intra-class difference is less than the intra-class threshold, and the false recognition rate is less than the false recognition rate threshold, to obtain updated facial features; and update the corresponding facial images in the facial image base library and the facial image test library based on the updated facial features.

10. The device according to claim 9, characterized in that The training module is also used to: Extracting facial feature vectors of each facial image in the facial image test library; Transmitting the extracted facial feature vector to a server, so that the server recognizes the facial feature vector using a facial recognition model to obtain a facial recognition prediction result; The triggering server compares the facial recognition prediction result with the corresponding facial recognition actual result to obtain a comparison result, and obtains an error recognition rate based on the comparison result; And, receive the error rate returned by the server.

11. The device according to claim 10, characterized in that The training module is also used to: detect key facial feature points in facial images and locate the key facial feature points; and extract the key facial feature points using a deep convolutional neural network to obtain a facial feature vector.

12. The device according to claim 10, characterized in that The server is used to obtain facial images in the facial image test library, and perform facial detection processing on the facial images to obtain facial images to be identified; perform facial alignment processing on the facial images to be identified to obtain facial images after alignment processing, and input the facial images after alignment processing into a facial recognition model, and the facial recognition model outputs facial recognition prediction results; and, based on the difference between the facial recognition prediction results and the corresponding facial recognition actual results, construct a loss function, and train the facial recognition model with the goal of minimizing the loss function, until the training is terminated when the training termination condition is met, and a trained facial recognition model is obtained.

13. The device according to claim 10, characterized in that The server is also used to calculate the similarity between the facial feature vector and the corresponding feature vector in the facial image base library; the server outputs the facial recognition prediction result according to the comparison result of the similarity and the similarity threshold.

14. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images