Construction method of face recognition data model, face recognition method and terminal

Abstract

The invention discloses a face recognition data model construction method, a face recognition method and a terminal, and relates to the technical field of face recognition, and the method comprises the steps: collecting multi-face images under different conditions through multiple types of cameras, carrying out graying after screening, carrying out Gaussian filtering, carrying out histogram equalization, carrying out YOLOv5 detection and positioning, constructing a CNN model based on TensorFlow, and extracting features. The method comprises the following steps: pre-training by using data sets such as LFW and the like, finely adjusting own data, training according to 6: 2: 2 score data, carrying out anti-overfitting on an SGD algorithm, a cross entropy loss function and an early stop method, and carrying out 100 times of iteration to obtain a model. The face recognition effect is improved, data model construction is comprehensive and accurate, the method is accurate and reliable, terminal functions are complete and practical, the recognition accuracy, speed and adaptability are improved, the application range is expanded, and the overall performance of the system and user experience are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of face recognition technology, and in particular to a method for constructing a face recognition data model, a face recognition method and a terminal. Background Art

[0002] With the rapid development of science and technology, face recognition technology has been widely used in many fields, such as security monitoring, access control systems, financial payment, social entertainment, etc. However, existing face recognition technology still faces many challenges and shortcomings.

[0003] In terms of data model construction, data collection is often not comprehensive and diverse enough. Many face recognition systems only collect data under limited lighting conditions, angles, and crowd ranges, resulting in poor adaptability of the model to different environments and individual characteristics. For example, in some low-light or strong direct light scenes, the accuracy of face recognition drops significantly; for non-frontal face images such as side faces, looking up or looking down, the recognition effect is also unsatisfactory. At the same time, there may be defects in the data preprocessing link, such as incomplete denoising and poor image contrast enhancement, which affect the accuracy of subsequent feature extraction. Feature extraction methods also need to be improved. Some traditional methods are difficult to effectively capture the subtle and deep features of the face, making the model prone to misjudgment when facing similar faces. In the process of model training, problems such as unreasonable division of training data, improper selection of training algorithms, and lack of effective measures to prevent overfitting are relatively common, resulting in insufficient generalization and stability of the model.

[0004] In the application of face recognition methods, the performance limitations of image acquisition equipment affect the quality of the image to be recognized. Some ordinary cameras have deficiencies in resolution, focus speed and anti-shake ability, which makes the collected face images blurred or deformed, increasing the difficulty of recognition. Moreover, in the face detection step, the existing detection algorithms are not accurate enough for faces in complex backgrounds, partially occluded faces or multi-face scenes, and are prone to missed detection or false detection. In terms of matching algorithms, a single similarity calculation method may not be able to accurately measure the differences in facial features, especially when the feature vector dimension is high and the data distribution is complex, which can easily lead to recognition errors. In addition, the lack of an effective secondary verification mechanism makes it impossible to further confirm the identity when the recognition result is ambiguous, reducing the reliability of the system.

[0005] On face recognition terminal devices, the functional integration is not perfect. The image acquisition module of some terminals lacks autofocus and anti-shake functions, and the image preprocessing and detection module has a slow calculation speed and cannot meet the real-time requirements. The model of the feature extraction and recognition module is not updated in a timely manner, and it is difficult to adapt to new face data features and application scenario changes. The result output module is not flexible enough and cannot provide personalized output format and content according to different user needs and application scenarios.

[0006] To sum up, the existing face recognition technology has many problems in data model construction, recognition methods and terminal equipment. There is an urgent need for a more advanced, reliable and adaptable face recognition technology solution to meet the growing market demand and application scenario requirements. Summary of the invention

[0007] The present invention proposes a method for constructing a face recognition data model, a face recognition method and a terminal to solve the problems mentioned in the above-mentioned prior art.

[0008] In order to achieve the above object, the present invention adopts the following technical solution: a method for constructing a face recognition data model, comprising the following steps: Data collection step: collect facial image data through image acquisition equipment, including high-definition cameras, depth cameras and infrared cameras, and photograph 1,000 individuals of different ages, genders and races under different lighting conditions and at different angles. The number of collected images is not less than 50,000, and the collected images are screened in real time to remove blurred and severely occluded images; Data preprocessing steps: Preprocess the collected face images. First, grayscale the images to convert the color images into grayscale images. Then, use the Gaussian filtering algorithm to remove image noise. The size of the Gaussian kernel is adjusted to 3x3 to balance the denoising effect and the degree of image detail retention. Use histogram equalization to enhance the image contrast and make the facial features clear. Perform face detection and positioning. Use the YOLOv5 target detection algorithm based on deep learning to identify the face area in the image and crop it out. Feature extraction steps: Use the deep learning framework TensorFlow to build a convolutional neural network (CNN) model for feature extraction. The CNN model contains 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. The convolutional layer extracts local features of the face image through convolution kernels of different sizes, including 3x3 or 5x5. The pooling layer performs downsampling operations to reduce the data dimension and retain the main features. The fully connected layer integrates and maps the extracted features to obtain a high-dimensional feature vector that represents the face. During the training process, public face datasets, including the LFW dataset and the CelebA dataset, are used to pre-train the CNN model, and then the collected own data is used for fine-tuning to meet the needs of face recognition tasks. Model training steps: The extracted facial feature data are divided into training set, validation set and test set in a ratio of 6:2:2; the training set is used to train the face recognition model, and the training algorithm adopts the stochastic gradient descent SGD algorithm. During the training process, a suitable loss function is set, including the cross entropy loss function, so that the loss function value gradually decreases, and the early stopping method EarlyStopping is used to prevent overfitting. When the loss function value on the validation set no longer decreases within 5 consecutive training cycles, the training is stopped; the performance of the model is evaluated on the test set, and the training parameters are adjusted according to the accuracy, recall rate, and F1 value indicators, including setting the learning rate to 0.001 and the batch size to 32. After 100 iterative training, a trained face recognition data model is obtained.

[0009] Furthermore, in the data collection step, 3D reconstruction technology is used to supplement the depth information of the collected facial images to enhance the model's ability to recognize facial shape features.

[0010] Furthermore, in the data preprocessing step, an adaptive histogram equalization algorithm is used to dynamically adjust the equalization parameters according to the local features of the image to improve the image contrast enhancement effect.

[0011] Furthermore, in the feature extraction step, an attention mechanism is introduced into the CNN model to make the model focus on the characteristic areas of the face and improve the quality of feature extraction.

[0012] Furthermore, in the model training step, an ensemble learning method is used to integrate the three trained face recognition models, and the overall performance and stability of the model are improved through weighted averaging or voting.

[0013] Furthermore, a method for performing face recognition using a face recognition data model obtained by the face recognition data model construction method includes the following steps: Steps for acquiring the image to be identified: The face image to be identified is acquired through an image acquisition device. The image acquisition device must ensure the clarity and resolution of the image. The acquired image format is a common image format, including JPEG and PNG. Image preprocessing and face detection steps: Preprocess the image to be recognized in a similar way to the data preprocessing step, including grayscale, denoising, and contrast enhancement operations; use the face detection algorithm to detect the face area in the image. If no face is detected or multiple faces are detected, give a corresponding prompt message and end the recognition process; Feature extraction and matching steps: Use the constructed face recognition data model to extract features from the detected face area to obtain the face feature vector of the image to be identified, and match the feature vector with the known face feature vectors stored in the database. The matching algorithm uses the cosine similarity algorithm to calculate the similarity between the feature vector to be identified and each feature vector in the database. The similarity threshold is set to 0.8. When the similarity exceeds the threshold, it is determined to be a successful match and the corresponding person’s identity is identified; if the similarity does not exceed the threshold, the recognition fails, the image to be identified is marked as an unknown person, and its feature vector can be selected to be stored in the database.

[0014] Furthermore, in the step of acquiring the image to be recognized, if the quality of the image acquired by the image acquisition device does not meet the requirements, the acquisition parameters, including focal length and exposure time, are automatically adjusted, and the image is re-acquired.

[0015] Furthermore, in the feature extraction and matching steps, when the matching result is in a fuzzy area near the similarity threshold, that is, when the similarity is between 0.7 and 0.8, a secondary verification mechanism is adopted, including combining liveness detection technology or requiring the input of additional identity authentication information, to improve the accuracy of the recognition result.

[0016] Furthermore, a face recognition terminal for executing the face recognition method includes: Image acquisition module: used to realize the function of acquiring the image to be recognized, including image acquisition devices such as high-definition cameras; Image preprocessing and detection module: performs image preprocessing and face detection steps, and performs preprocessing and face detection operations on the collected images; Feature extraction and recognition module: according to the feature extraction and matching steps, the constructed face recognition data model is used to perform feature extraction and matching recognition; Result output module: outputs the results of face recognition by displaying them on the terminal screen, broadcasting them by voice, or transmitting them to other devices via the network.

[0017] Furthermore, the image acquisition module has auto-focus and image anti-shake functions, ensuring that high-quality images can be acquired in different environments.

[0018] The image preprocessing and detection module adopts hardware acceleration technology, including GPU acceleration; The feature extraction and recognition module has a model update function, and regularly downloads updated face recognition data models from the server; The result output module customizes the output format and content according to user requirements to meet the needs of different application scenarios.

[0019] Compared with the prior art, the present invention has the following beneficial effects: This patent has significant beneficial effects in many aspects. In data model construction, a large number of diverse facial images are collected through multiple types of cameras, covering different lighting, angles and crowds, ensuring rich and comprehensive data. Advanced data preprocessing technology effectively removes noise, enhances contrast and accurately locates faces, laying a good foundation for subsequent steps. The convolutional neural network built with a deep learning framework combined with an attention mechanism can accurately extract key facial features. The model obtained by reasonably dividing the data set and optimizing the training algorithm has strong generalization ability and high stability, and can adapt to complex and changeable face recognition tasks.

[0020] In the face recognition method, the image acquisition device can automatically optimize the acquisition parameters to obtain high-quality images, and the face detection algorithm can accurately identify in complex scenes and reduce the false detection rate. The cosine similarity algorithm is used with a reasonable threshold, combined with a secondary verification mechanism, and the use of liveness detection or additional information verification in fuzzy areas greatly improves the recognition accuracy and reliability, and effectively avoids the occurrence of false recognition.

[0021] In terms of face recognition terminals, the autofocus and anti-shake functions of the image acquisition module ensure image quality, and the hardware acceleration technology of the image preprocessing and detection module achieves fast processing to meet real-time requirements. The model update function of the feature extraction and recognition module enables it to keep up with technological development and data changes, and continuously improve recognition performance. The result output module can customize the output to meet the needs of different scenarios, such as detailed alarm information output in the security field, and interesting interactive results display in social entertainment applications, which greatly expands the application scope and user experience of face recognition technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic block diagram of a method for constructing a face recognition data model proposed by the present invention; Figure 2 The present invention is a schematic block diagram of a face recognition method proposed by the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0025] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, and it can be the internal connection of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0026] Reference Figure 1-2 :A method for constructing a face recognition data model, comprising the following steps: Data collection steps: Collect facial image data through a variety of image acquisition devices, including high-definition cameras, depth cameras, and infrared cameras. Under different lighting conditions, such as strong light, weak light, natural light, and different angles, including front, side, upward, and downward, 1,000 individuals of different ages, genders, and races are photographed, and the number of collected images is no less than 50,000 to ensure the diversity and comprehensiveness of the data. At the same time, the collected images are screened in real time to remove blurred and severely blocked images and retain clear images with complete facial features.

[0027] Data preprocessing steps: Preprocess the collected face images. First, grayscale the image to convert the color image into a grayscale image to reduce the amount of data and computational complexity. Then, use the Gaussian filter algorithm to remove image noise, and adjust the size of the Gaussian kernel to 3x3 to balance the denoising effect and the degree of image detail retention. Next, use histogram equalization to enhance the image contrast and make the facial features more obvious. After that, perform face detection and positioning, and use a deep learning-based target detection algorithm, such as the YOLOv5 algorithm, to accurately identify the face area in the image and crop it out for subsequent processing.

[0028] Feature extraction step: Using the deep learning framework TensorFlow, a convolutional neural network (CNN) model is constructed for feature extraction. The CNN model contains 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. The convolutional layer extracts local features of the face image through convolution kernels of different sizes, such as 3x3, 5x5, etc. The pooling layer performs downsampling operations to reduce the data dimension and retain the main features. The fully connected layer integrates and maps the previously extracted features to obtain a high-dimensional feature vector that can represent the face. During the training process, large-scale public face datasets, such as the LFW dataset and the CelebA dataset, are used to pre-train the CNN model, and then the collected own data is used for fine-tuning to adapt to the specific face recognition task requirements and improve the accuracy and effectiveness of feature extraction.

[0029] Model training steps: Divide the extracted facial feature data into training set, validation set and test set in a ratio of 6:2:2. Use the training set to train the face recognition model. The training algorithm adopts the stochastic gradient descent (SGD) algorithm. During the training process, set a suitable loss function, such as the cross entropy loss function, and gradually reduce the loss function value by continuously adjusting the parameters of the model. At the same time, the early stopping method (EarlyStopping) is used to prevent overfitting, that is, when the loss function value on the validation set no longer decreases within 5 consecutive training cycles, stop training. Regularly evaluate the performance of the model on the test set, and adjust the training parameters according to indicators such as accuracy, recall rate, F1 value, etc., such as setting the learning rate to 0.001 and the batch size to 32. After 100 iterations of training, a trained face recognition data model is obtained.

[0030] In the data collection step, the present invention uses 3D reconstruction technology to supplement the depth information of the collected face image, thereby enhancing the model's recognition ability of face shape features.

[0031] The present invention adopts an adaptive histogram equalization algorithm in the data preprocessing step to dynamically adjust the equalization parameters according to the local features of the image, thereby further improving the image contrast enhancement effect.

[0032] In the feature extraction step, the present invention introduces an attention mechanism into the CNN model, so that the model can focus on the key feature areas of the face and improve the quality of feature extraction.

[0033] In the model training step, the present invention adopts an ensemble learning method to integrate three trained face recognition models, such as through weighted averaging or voting, to improve the overall performance and stability of the model.

[0034] The present invention also discloses a face recognition method, comprising the following steps: Steps for acquiring the image to be identified: The face image to be identified is acquired through an image acquisition device. The image acquisition device must ensure the clarity and resolution of the image. The acquired image format is a common image format, such as JPEG, PNG, etc.

[0035] Image preprocessing and face detection step: The image to be identified is preprocessed in a manner similar to the data preprocessing step in claim 1, including grayscale, denoising, contrast enhancement, etc. Then, a face detection algorithm is used to detect the face area in the image. If no face is detected or multiple faces are detected, a corresponding prompt message is given and the identification process ends.

[0036] Feature extraction and matching step: Use the face recognition data model constructed in claim 1 to extract features from the detected face area to obtain the face feature vector of the image to be identified. Match the feature vector with the known face feature vectors stored in the database, and the matching algorithm adopts the cosine similarity algorithm. Calculate the similarity between the feature vector to be identified and each feature vector in the database, set the similarity threshold to 0.8, and when the similarity exceeds the threshold, it is determined to be a successful match and the corresponding person’s identity is identified; if the similarity does not exceed the threshold, the recognition fails, and the image to be identified is marked as an unknown person, and its feature vector can be selected to be stored in the database for subsequent manual confirmation or further analysis.

[0037] In the step of acquiring the image to be recognized, if the quality of the image acquired by the image acquisition device does not meet the requirements, the acquisition parameters, such as focal length, exposure time, etc., are automatically adjusted to re-acquire the image.

[0038] In the feature extraction and matching steps of the present invention, when the matching result is in a fuzzy area near the similarity threshold, that is, when the similarity is between 0.7 and 0.8, a secondary verification mechanism is adopted, such as combining liveness detection technology or requiring input of additional identity authentication information, to improve the accuracy of the recognition result.

[0039] The present invention also discloses a face recognition terminal, comprising: Image acquisition module: used to implement the function of the step of acquiring the image to be identified in claim 6, including image acquisition devices such as high-definition cameras, which can acquire clear facial images.

[0040] Image preprocessing and detection module: executes the image preprocessing and face detection steps in claim 6, and performs preprocessing and face detection operations on the collected images.

[0041] Feature extraction and recognition module: according to the feature extraction and matching steps in claim 6, feature extraction and matching recognition are performed using the constructed face recognition data model.

[0042] Result output module: outputs the results of face recognition. The output can be displayed on the terminal screen, broadcast by voice, or transmitted to other devices through the network.

[0043] In the present invention, the image acquisition module has automatic focusing and image anti-shake functions, ensuring that high-quality images can be acquired in different environments.

[0044] In the present invention, the image preprocessing and detection module adopts hardware acceleration technology, such as GPU acceleration, to improve the speed of image preprocessing and face detection.

[0045] In the present invention, the feature extraction and recognition module has a model updating function and can regularly download updated face recognition data models from a server to improve recognition performance.

[0046] In the present invention, the result output module can customize the output format and content according to user needs to meet the needs of different application scenarios.

[0047] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for constructing a face recognition data model, characterized in that: The following steps are involved: Data collection step: collect facial image data through image acquisition equipment, including high-definition cameras, depth cameras and infrared cameras, and photograph 1,000 individuals of different ages, genders and races under different lighting conditions and at different angles. The number of collected images is not less than 50,000, and the collected images are screened in real time to remove blurred and severely occluded images; Data preprocessing steps: Preprocess the collected face images. First, grayscale the images to convert the color images into grayscale images. Then, use the Gaussian filtering algorithm to remove image noise. The size of the Gaussian kernel is adjusted to 3x3 to balance the denoising effect and the degree of image detail retention. Use histogram equalization to enhance the image contrast and make the facial features clear. Perform face detection and positioning. Use the YOLOv5 target detection algorithm based on deep learning to identify the face area in the image and crop it out. Feature extraction steps: Use the deep learning framework TensorFlow to build a convolutional neural network (CNN) model for feature extraction. The CNN model contains 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. The convolutional layer extracts local features of the face image through convolution kernels of different sizes, including 3x3 or 5x5. The pooling layer performs downsampling operations to reduce the data dimension and retain the main features. The fully connected layer integrates and maps the extracted features to obtain a high-dimensional feature vector that represents the face. During the training process, public face datasets, including the LFW dataset and the CelebA dataset, are used to pre-train the CNN model, and then the collected own data is used for fine-tuning to meet the needs of face recognition tasks. Model training steps: The extracted facial feature data are divided into training set, validation set and test set in a ratio of 6:2:2; the training set is used to train the face recognition model, and the training algorithm adopts the stochastic gradient descent SGD algorithm. During the training process, a suitable loss function is set, including the cross entropy loss function, so that the loss function value gradually decreases, and the early stopping method EarlyStopping is used to prevent overfitting. When the loss function value on the validation set no longer decreases within 5 consecutive training cycles, the training is stopped; the performance of the model is evaluated on the test set, and the training parameters are adjusted according to the accuracy, recall rate, and F1 value indicators, including setting the learning rate to 0.001 and the batch size to 32. After 100 iterative training, a trained face recognition data model is obtained.

2. The method for constructing a face recognition data model according to claim 1, characterized in that: In the data collection step, 3D reconstruction technology is used to supplement the depth information of the collected facial images to enhance the model's ability to recognize facial shape features.

3. The method for constructing a face recognition data model according to claim 1, characterized in that: In the data preprocessing step, an adaptive histogram equalization algorithm is used to dynamically adjust the equalization parameters according to the local characteristics of the image to improve the image contrast enhancement effect.

4. The method for constructing a face recognition data model according to claim 1, characterized in that: In the feature extraction step, the attention mechanism is introduced into the CNN model to make the model focus on the characteristic areas of the face and improve the quality of feature extraction.

5. The method for constructing a face recognition data model according to claim 1, characterized in that: In the model training step, an ensemble learning method is used to integrate the three trained face recognition models, and the overall performance and stability of the model are improved through weighted averaging or voting.

6. A method for face recognition using a face recognition data model obtained by the face recognition data model construction method according to any one of claims 1 to 5, characterized in that: The following steps are involved: Steps for acquiring the image to be identified: The face image to be identified is acquired through an image acquisition device. The image acquisition device must ensure the clarity and resolution of the image. The acquired image format is a common image format, including JPEG and PNG. Image preprocessing and face detection steps: Preprocess the image to be recognized in a similar way to the data preprocessing step, including grayscale, denoising, and contrast enhancement operations; use the face detection algorithm to detect the face area in the image. If no face is detected or multiple faces are detected, give a corresponding prompt message and end the recognition process; Feature extraction and matching steps: Use the constructed face recognition data model to extract features from the detected face area to obtain the face feature vector of the image to be identified, and match the feature vector with the known face feature vectors stored in the database. The matching algorithm uses the cosine similarity algorithm to calculate the similarity between the feature vector to be identified and each feature vector in the database. The similarity threshold is set to 0.

8. When the similarity exceeds the threshold, it is determined to be a successful match and the corresponding person’s identity is identified; if the similarity does not exceed the threshold, the recognition fails, the image to be identified is marked as an unknown person, and its feature vector can be selected to be stored in the database.

7. A method for face recognition according to claim 6, characterized in that: In the step of acquiring the image to be recognized, if the image quality acquired by the image acquisition device does not meet the requirements, the acquisition parameters, including focal length and exposure time, are automatically adjusted and the image is re-acquired.

8. A face recognition method according to claim 6, characterized in that: In the feature extraction and matching steps, when the matching result is in the fuzzy area near the similarity threshold, that is, the similarity is between 0.7 and 0.8, a secondary verification mechanism is adopted, including combining liveness detection technology or requiring the input of additional identity authentication information, to improve the accuracy of the recognition result.

9. A face recognition terminal for executing the face recognition method according to any one of claims 6 to 8, characterized in that: include: Image acquisition module: used to realize the function of acquiring the image to be recognized, including image acquisition devices such as high-definition cameras; Image preprocessing and detection module: performs image preprocessing and face detection steps, and performs preprocessing and face detection operations on the collected images; Feature extraction and recognition module: according to the feature extraction and matching steps, the constructed face recognition data model is used to perform feature extraction and matching recognition; Result output module: outputs the results of face recognition by displaying them on the terminal screen, broadcasting them by voice, or transmitting them to other devices via the network.

10. A face recognition terminal according to claim 9, characterized in that: The image acquisition module has auto-focus and image anti-shake functions to ensure that high-quality images can be acquired in different environments; The image preprocessing and detection module adopts hardware acceleration technology, including GPU acceleration; The feature extraction and recognition module has a model update function, and regularly downloads updated face recognition data models from the server; The result output module customizes the output format and content according to user requirements to meet the needs of different application scenarios.

Images