Human body image processing method and device based on clustering technology, equipment and medium

Abstract

The application discloses a human body image processing method and device based on clustering technology, equipment and medium, relates to the field of processor architecture design, and comprises the following steps: obtaining an image processing instruction of a target client and an initial human body image sent by a host; determining a target feature parameter based on the initial human body image; determining a target image feature by using the target feature parameter and the initial human body image; determining a clustering number based on the image processing instruction and the target image feature by using a local state machine, determining a target clustering algorithm based on the image processing instruction, and clustering the initial human body image by using the target clustering algorithm, the clustering number and the target image feature to obtain a corresponding clustering result; determining each human body image set based on the clustering result, determining an image allocation scheduling strategy, performing image processing on each human body image set by using the image allocation scheduling strategy, and sending the obtained image processing result to the target client. Therefore, the application can improve the efficiency of human body image processing.

Description

Technical Field

[0001] This invention relates to the field of processor architecture design, and in particular to a method, apparatus, device, and medium for human image processing based on clustering technology. Background Technology

[0002] In modern computer systems, the processor, as a core component, encompasses both the CPU (Central Processing Unit) and the GPU (Graphics Processing Unit), primarily responsible for parsing instructions and rapidly processing complex data. With the rapid development of technology, the demand for computational performance from dedicated image processors is increasing. Researchers have proposed introducing clustering techniques into image processing to improve efficiency and accuracy by classifying and grouping massive amounts of images according to key features.

[0003] However, in the field of human image processing, existing clustering-based solutions suffer from several significant problems in practice. First, traditional architectures rely on the host CPU to perform critical human image clustering calculations, grouping images, allocating processing units, and scheduling tasks based on the results. This places an extremely high load on the CPU, leading to excessive latency in other important processes and severely impacting user experience. Second, existing solutions often employ an architecture that separates clustering calculations from the processor's image processing unit. User instructions and image data to be processed must first undergo clustering algorithm execution by the host CPU, and the results are then scheduled and transmitted to the processor via the on-chip bus for processing, significantly reducing the communication efficiency of critical human image processing processes. Furthermore, existing solutions offer a limited variety of deployable clustering algorithms, making it impossible to select the most suitable algorithm based on the specific human image, thus affecting the clustering results. Simultaneously, existing architectures are complex, requiring highly sophisticated hardware and software collaboration. Users need to master complex programming and development fundamentals, greatly hindering the application and development of deep learning systems by researchers.

[0004] Therefore, improving the efficiency, accuracy, and convenience of human body image processing is a pressing technical problem that needs to be solved. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for human image processing based on clustering technology, which can improve the efficiency, accuracy, and convenience of human image processing. The specific solution is as follows:

[0006] In a first aspect, this application provides a human image processing method based on clustering technology, applied to an image processor, comprising:

[0007] The system acquires the image processing instructions and corresponding initial human body image to be processed from the target client sent by the host; the initial human body image is an image including a face and / or human body movements.

[0008] The initial human image is cleaned, an HSV image is determined based on the cleaned image, and a hash operation is performed on the key point feature descriptors determined from the HSV image to determine the corresponding target feature parameters from a preset feature parameter storage table using the obtained target hash address; the preset feature parameter storage table is a table that records the correspondence between the feature parameters of the human image and the hash address.

[0009] Based on a preset convolutional neural network model, the initial image features of the initial human body image are determined using the target feature parameters and the initial human body image, and the initial image features are standardized to obtain the target image features.

[0010] The initial human image is clustered using a local state machine, based on the cleaned image, the image processing instructions, and the target image features. The number of clusters is determined, the target clustering algorithm is determined based on the image processing instructions, and the initial human image is clustered using the target clustering algorithm, the number of clusters, the cleaned image, and the target image features to obtain the corresponding clustering results.

[0011] Based on the clustering results, each human image set is determined by image category. Based on the image processing instructions, an image allocation and scheduling strategy for each human image set is determined. The image allocation and scheduling strategy is used to allocate each human image set to each image processing component for image processing, so that the obtained image processing results are sent to the target client through the host.

[0012] Optionally, obtaining the image processing instructions for the target client sent by the host and the corresponding initial human image to be processed includes:

[0013] The image processor acquires the image processing instructions and the corresponding initial human body image to be processed from the target client sent by the host through a preset interface and based on direct memory access, so that the image processor saves the acquired image processing instructions and the initial human body image to a preset image cache component in the image processor.

[0014] Optionally, the cleaning of the initial human image includes:

[0015] Determine the image quality of the initial human body image;

[0016] In the initial human body image, human body images that meet the preset low quality conditions are removed by using preset image cleaning parameters configured on the target user terminal, and the brightness and contrast of the initial human body image after image removal are adjusted to complete the image cleaning operation of the initial human body image and obtain the corresponding cleaned image.

[0017] Optionally, the step of determining the HSV image based on the obtained cleaned image and performing a hash operation on the key point feature descriptors determined from the HSV image to determine the corresponding target feature parameters from a preset feature parameter storage table using the obtained target hash address includes:

[0018] The cleaned image is then resized and the resulting resized image is converted to a color space to obtain an HSV image.

[0019] Select any one of the hue, saturation, and brightness channels in the HSV image as the target channel, and use a preset key point detector to determine key point feature descriptors from the target channel of the HSV image, and perform a hash operation on the key point feature descriptors to obtain the target hash address;

[0020] The target feature parameters corresponding to the target hash address are determined from the preset feature parameter storage table.

[0021] Optionally, the standardization process for the initial image features to obtain the target image features includes:

[0022] The initial image features are subjected to Z-Score normalization and preset feature dimensionality reduction to obtain the corresponding processed image features;

[0023] The processed image features are subjected to a preset feature fusion operation to obtain target image features, which are then sent to the state machine of the image processor.

[0024] Optionally, the algorithm for determining the target clustering based on the image processing instructions includes:

[0025] Based on the image processing instructions, determine the feature dimension, feature quantity, and noise anomaly ratio of the target image features;

[0026] By comparing the feature dimension with a preset dimension threshold, a first comparison result is obtained;

[0027] If the first comparison result indicates that the feature dimension is not greater than the preset dimension threshold, then the K-Means clustering algorithm is determined as the target clustering algorithm;

[0028] If the first comparison result indicates that the feature dimension is greater than the preset dimension threshold, then the spectral clustering algorithm is determined as the target clustering algorithm;

[0029] By comparing the feature quantity with a preset quantity threshold, a second comparison result is obtained;

[0030] If the second comparison result indicates that the feature quantity is less than the preset quantity threshold, then the spectral clustering algorithm is determined as the target clustering algorithm;

[0031] If the second comparison result indicates that the feature quantity is not less than the preset quantity threshold, then the K-Means clustering algorithm is determined as the target clustering algorithm;

[0032] By comparing the noise anomaly ratio with the preset ratio threshold, a third comparison result is obtained;

[0033] If the third comparison result indicates that the proportion of noise anomalies is greater than the preset proportion threshold, then the DBSCAN clustering algorithm is determined as the target clustering algorithm.

[0034] If the third comparison result indicates that the proportion of noise anomalies is not greater than the preset proportion threshold, then the spectral clustering algorithm or the K-Means clustering algorithm is determined as the target clustering algorithm.

[0035] Optionally, determining the image allocation and scheduling strategy for each of the human image sets based on the image processing instructions includes:

[0036] The image processing complexity is determined based on the image processing instructions.

[0037] Based on the image processing complexity, the type and number of image processing components corresponding to each of the human body image sets are determined, and the image allocation and scheduling strategy for each of the human body image sets is determined using the type and the number.

[0038] Secondly, this application provides a human image processing device based on clustering technology, applied to an image processor, comprising:

[0039] The information acquisition module is used to acquire the image processing instructions and the corresponding initial human body image to be processed from the target client sent by the host; the initial human body image is an image including a face and / or human body movements;

[0040] The parameter determination module is used to clean the initial human body image, determine an HSV image based on the cleaned image, and perform a hash operation on the key point feature descriptors determined from the HSV image, so as to determine the corresponding target feature parameters from the preset feature parameter storage table using the obtained target hash address; the preset feature parameter storage table is a table that records the correspondence between the feature parameters of the human body image and the hash address.

[0041] The feature processing module is used to determine the initial image features of the initial human image based on a preset convolutional neural network model and using the target feature parameters and the initial human image, and to standardize the initial image features to obtain the target image features;

[0042] The image clustering module is used to determine the number of clusters based on the cleaned image, the image processing instructions, and the target image features through a local state machine, determine the target clustering algorithm based on the image processing instructions, and cluster the initial human image using the target clustering algorithm, the number of clusters, the cleaned image, and the target image features to obtain the corresponding clustering results.

[0043] The image processing module is used to determine each human image set grouped by image category based on the clustering results, determine the image allocation and scheduling strategy for each human image set based on the image processing instructions, allocate each human image set to each image processing component for image processing using the image allocation and scheduling strategy, and send the obtained image processing results to the target client through the host.

[0044] Thirdly, this application provides an electronic device, comprising:

[0045] Memory, used to store computer programs;

[0046] A processor is used to execute the computer program to implement the aforementioned human image processing method based on clustering technology.

[0047] Fourthly, this application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned human image processing method based on clustering technology.

[0048] In this application, an image processing instruction from a target client sent by a host and a corresponding initial human image to be processed are obtained; the initial human image is an image including a face and / or human movements; the initial human image is cleaned, an HSV image is determined based on the cleaned image, and a hash operation is performed on the key point feature descriptors determined from the HSV image to determine the corresponding target feature parameters from a preset feature parameter storage table using the obtained target hash address; the preset feature parameter storage table is a table that records the correspondence between the feature parameters of the human image and the hash address; based on a preset convolutional neural network model and using the target feature parameters and the initial human image, the initial image features of the initial human image are determined, and the initial image features are labeled. The system performs standardization processing to obtain target image features. Using a local state machine, and based on the cleaned image, the image processing instructions, and the target image features, it determines the number of clusters, determines a target clustering algorithm based on the image processing instructions, and uses the target clustering algorithm, the number of clusters, the cleaned image, and the target image features to cluster the initial human image, obtaining corresponding clustering results. Based on the clustering results, it determines each human image set grouped by image category, determines an image allocation and scheduling strategy for each human image set based on the image processing instructions, and uses the image allocation and scheduling strategy to allocate each human image set to each image processing component for image processing, so that the obtained image processing results are sent to the target client through the host. As can be seen from the above, in this application, the image processing instructions of the target client sent by the host and the corresponding initial human body image to be processed are received first; then, a cleaning operation is performed on the initial human body image, and an HSV image is determined based on the image obtained after cleaning. Then, a hash operation is performed on the key point feature descriptors extracted from the HSV image, and the corresponding target feature parameters are searched and determined from the preset feature parameter storage table based on the obtained target hash address; then, with the help of a preset convolutional neural network model, the initial image features of the initial human body image are determined by combining the target feature parameters and the initial human body image. Subsequently, the initial image features are standardized to finally obtain the target image features; through the local state machine, the number of clusters is determined according to the cleaned image, the image processing instructions and the target image features. At the same time, a target clustering algorithm is selected according to the image processing instructions. Then, the target clustering algorithm, the determined number of clusters, the cleaned image and the target image features are used to perform clustering operations on the initial human body image to obtain the corresponding clustering results.Finally, based on the clustering results, various human image sets are grouped according to image categories. An image allocation and scheduling strategy for each of these human image sets is formulated based on the image processing instructions. This strategy is then used to assign each human image set to a specific image processing component for processing. Finally, the processed image results are sent to the target client via the host computer. In this way, this application can improve the efficiency, accuracy, and convenience of human image processing, and to a certain extent, enhance the user experience. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the system architecture of a high-efficiency image processor based on clustering technology disclosed in this application;

[0051] Figure 2 This is a flowchart of a human image processing method based on clustering technology disclosed in this application;

[0052] Figure 3 This application discloses a specific flowchart of a human image processing method based on clustering technology.

[0053] Figure 4 This is a schematic diagram of the architecture of a clustering preprocessing unit disclosed in this application;

[0054] Figure 5 This is a schematic diagram of an image clustering preprocessing process for a clustering preprocessing unit disclosed in this application;

[0055] Figure 6 This is a flowchart of a clustering algorithm selection method disclosed in this application;

[0056] Figure 7 This is a schematic diagram of a human image processing device based on clustering technology disclosed in this application;

[0057] Figure 8 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Currently, existing human image processing solutions based on clustering technology suffer from several significant problems in practice. First, traditional architectures rely on the host CPU to perform critical human image clustering calculations, grouping images, allocating processing units, and scheduling tasks based on the results. This places an extremely high load on the CPU, leading to excessive latency in other important processes and severely impacting user experience. Second, existing solutions often employ an architecture that separates clustering calculations from the processor's image processing unit. User instructions and image data to be processed must first undergo clustering algorithm execution by the host CPU, and the results are then scheduled and transmitted to the processor's processing unit via the on-chip bus for further processing, significantly reducing the communication efficiency of critical human image processing processes. Furthermore, existing solutions offer a limited variety of deployable clustering algorithms, making it impossible to select the most suitable algorithm based on the specific human image, thus affecting the clustering results. Simultaneously, existing architectures are complex, requiring highly sophisticated hardware and software collaboration, and demanding that users possess complex programming and development skills, greatly hindering the application and development of deep learning systems by researchers. Therefore, this application provides a human image processing method, apparatus, device, and medium based on clustering technology, which can improve the efficiency, accuracy, and convenience of human image processing.

[0060] The image processing system architecture involved in the human image processing method based on clustering technology in this application can be as follows: Figure 1 As shown.

[0061] See Figure 1 As shown, the system architecture includes a host, an image clustering preprocessing module, a DMA (Direct Memory Access) data transceiver module, a graph data processing unit control module, an on-chip cache module, and an AXI control / data bus connecting the various modules. The image clustering preprocessing module contains a clustering preprocessing unit, an image cache unit, and an image task scheduling unit; the graph data processing unit control module contains all image data processing units deployed with the high-efficiency image processor; and the on-chip cache module contains an instruction TCM (Instruction Tightly Coupled Memory) and a data TCM (Data Tightly Coupled Memory).

[0062] It should be noted that the Host is the control core of the system, responsible for coordinating and scheduling the user's image processing tasks. The system components control and complete the configuration of the relevant system parameters. The DMA data transceiver module deploys a DMA data transceiver engine, which is responsible for data exchange between the processor and the Host memory, and for sending and receiving computing task-related information and transmitting computing results in real time.

[0063] The image clustering preprocessing module includes an image clustering preprocessing unit, an image feature parameter storage table (i.e., a preset feature parameter storage table that stores all possible user-input image feature parameter information), an image caching module, and an image processing task scheduling module. The image clustering preprocessing module is responsible for performing a series of clustering preprocessing operations on the image data to be processed based on the current user's image processing request, such as feature extraction, feature standardization, and clustering algorithm execution. After completing the corresponding clustering and grouping of the image, it sends the specific image data to be processed to the graph data processing unit control module. The image clustering preprocessing module and the graph data processing unit control module adopt an integrated design, allowing direct task scheduling of the image after clustering calculation and grouping, without the need for cumbersome and time-consuming cache relay communication.

[0064] The image data processing unit control module deploys all the image processing units of the processor, and is responsible for executing image data processing tasks to complete clustering and grouping. The image processing unit includes the necessary scalar data calculation functions, vector data calculation functions, tensor data calculation functions, etc.

[0065] The on-chip cache includes instruction TCM storage and data TCM storage, which is responsible for storing relevant instruction data and calculating raw input data, and caching image processing results.

[0066] In addition, different modules are connected via an AXI (Advanced eXtensible Interface) bus, which is responsible for transmitting the instructions and related data necessary for the operation of each module. All functional modules are uniformly controlled and deployed by a dedicated image processor based on clustering technology.

[0067] Based on the above image processing architecture, it can be understood that the general image processing flow of this application is as follows: The server user node connects to the Host via a wired or wireless network. The heterogeneous high-efficiency image processor communicates and exchanges information with the Host in the form of DMA through the PCIe (Peripheral Component Interconnect Express) interface. After receiving the user's image processing request, the Host sends the calculation instructions (i.e., image processing instructions) and the image data to be processed (i.e., the initial human image) to the on-chip cache of the dedicated image processor based on clustering technology through DMA. The image clustering preprocessing module of the high-efficiency image processor calculates the grouping of all images to be processed in sequence according to the input processing instructions and the relevant user-set clustering configuration data, and schedules the specific calculation tasks to the target calculation unit of the image data processing unit control module. After the corresponding calculation unit completes the calculation, the calculation results are temporarily stored in the on-chip cache, and then the calculation results are sent to the Host through the PCIe interface in the form of DMA, and then fed back to the requesting user terminal.

[0068] See Figure 2 As shown, this embodiment of the invention discloses a human image processing method based on clustering technology, applied to an image processor, including:

[0069] Step S11: Obtain the image processing instructions and corresponding initial human body image to be processed from the target client sent by the host; the initial human body image is an image including a face and / or human body movements.

[0070] In this embodiment, the image processing instructions and the corresponding initial human image to be processed are obtained from the target client sent by the host through a preset interface and based on direct memory access. The image processor then saves the obtained image processing instructions and the initial human image to a preset image cache component within the image processor. Subsequently, clustering-related processing operations are performed on the initial human image based on these instructions. The initial human image can be image data containing key human information, acquired by the target client through its own camera or retrieved from local storage. Specifically, it can be a portrait image showing only facial features or a scene image showing human limb movements.

[0071] Step S12: Clean the initial human image, determine the HSV image based on the cleaned image, and perform a hash operation on the key point feature descriptors determined from the HSV image to determine the corresponding target feature parameters from the preset feature parameter storage table using the obtained target hash address; the preset feature parameter storage table is a table that records the correspondence between the feature parameters of the human image and the hash address.

[0072] In this embodiment, after acquiring the initial human body image, the image processor first performs an image cleaning operation, and image quality assessment is a prerequisite for the cleaning operation. The image quality of the initial human body image needs to be determined first. In the initial human body image, human body images that meet preset low-quality conditions are removed using preset image cleaning parameters configured on the target user terminal. The brightness and contrast of the initial human body image after image removal are then adjusted to complete the image cleaning operation and obtain the corresponding cleaned image.

[0073] After image cleaning, the cleaned image is resized and then converted to an HSV (Hue, Saturation, Value) image. Since the HSV space more intuitively reflects the hue, vibrancy, and brightness of colors, and each channel is relatively independent, it is more suitable for subsequent feature extraction operations.

[0074] Next, any one of the hue, saturation, and brightness channels in the HSV image is selected as the target channel, and a preset key point detector is used to determine the key point feature descriptor from the target channel of the HSV image. The key point feature descriptor is then hashed to obtain the target hash address.

[0075] Simultaneously, the image processor accesses a pre-stored feature parameter storage table. This table, established through pre-training, records the mapping relationship between feature parameters of numerous human images and their corresponding hash addresses. The target feature parameters corresponding to the target hash address are then determined from the pre-stored feature parameter storage table, providing foundational data support for subsequent human image analysis based on clustering techniques.

[0076] Step S13: Based on a preset convolutional neural network model and using the target feature parameters and the initial human image, determine the initial image features of the initial human image, and perform standardization processing on the initial image features to obtain the target image features.

[0077] In this embodiment, the image processor invokes a preset convolutional neural network model, which extracts features from the initial human image and outputs the initial image features. Next, the initial image features are subjected to Z-score normalization and preset feature dimensionality reduction to obtain the corresponding processed image features. Finally, the processed image features are subjected to a preset feature fusion operation to obtain the target image features, which are then sent to the state machine of the image processor to provide standardized input data for subsequent clustering steps.

[0078] Step S14: Using a local state machine, determine the number of clusters based on the cleaned image, the image processing instructions, and the target image features; determine the target clustering algorithm based on the image processing instructions; and use the target clustering algorithm, the number of clusters, the cleaned image, and the target image features to cluster the initial human image to obtain the corresponding clustering results.

[0079] In this embodiment, the feature dimension, feature quantity, and noise anomaly ratio of the target image features are determined by the local state machine based on the image processing instructions.

[0080] The feature dimension is compared with a preset dimension threshold to obtain a first comparison result. If the first comparison result indicates that the feature dimension is not greater than the preset dimension threshold, then the K-Means clustering algorithm is determined as the target clustering algorithm; if the first comparison result indicates that the feature dimension is greater than the preset dimension threshold, then the spectral clustering algorithm is determined as the target clustering algorithm.

[0081] The feature quantity is compared with a preset quantity threshold to obtain a second comparison result. If the second comparison result indicates that the feature quantity is less than the preset quantity threshold, then the spectral clustering algorithm is determined as the target clustering algorithm; if the second comparison result indicates that the feature quantity is not less than the preset quantity threshold, then the K-Means clustering algorithm is determined as the target clustering algorithm.

[0082] A third comparison result is obtained by comparing the noise anomaly ratio with a preset ratio threshold. If the third comparison result indicates that the noise anomaly ratio is greater than the preset ratio threshold, then the DBSCAN clustering algorithm is determined as the target clustering algorithm; if the third comparison result indicates that the noise anomaly ratio is not greater than the preset ratio threshold, then the spectral clustering algorithm or the K-Means clustering algorithm is determined as the target clustering algorithm.

[0083] After determining the target clustering algorithm, the state machine further calculates and determines the number of clusters based on the feature distribution of the cleaned image, image processing instructions, and target image features. Finally, the state machine calls the selected target clustering algorithm, using the determined number of clusters as parameters, and combines the features of the cleaned image and the target image to perform clustering operations on the initial human images. During the clustering process, the target clustering algorithm can calculate the feature similarity between samples, grouping human images with similar features into the same category, and finally outputting a clustering result containing multiple clusters, with each cluster corresponding to a group of human images with similar features.

[0084] Step S15: Based on the clustering results, determine each human image set grouped by image category; based on the image processing instructions, determine the image allocation and scheduling strategy for each human image set; use the image allocation and scheduling strategy to allocate each human image set to each image processing component for image processing; and send the obtained image processing results to the target client through the host.

[0085] In this embodiment, after obtaining the clustering results, the image processor determines each human image set grouped by image category, and each image set may contain human images with similar features.

[0086] Subsequently, the image processing complexity is determined based on the image processing instructions, and the type and quantity of image processing components corresponding to each of the human image sets are determined based on the image processing complexity. The image allocation and scheduling strategy for each human image set is then determined using the type and quantity. That is, the image processor matches the type and required quantity of image processing components corresponding to each human image set. Based on this, an image allocation and scheduling strategy is generated. The image processor distributes each human image set to the corresponding image processing component according to the image allocation and scheduling strategy. After the component completes processing, it feeds back the results. Finally, the image processor summarizes all results and sends them to the target client through the host, completing the entire human image processing flow based on clustering technology.

[0087] As can be seen from the above, in this application, the image processing instructions of the target client sent by the host and the corresponding initial human body image to be processed are received first; then, a cleaning operation is performed on the initial human body image, and an HSV image is determined based on the image obtained after cleaning. Then, a hash operation is performed on the key point feature descriptors extracted from the HSV image, and the corresponding target feature parameters are searched and determined from the preset feature parameter storage table based on the obtained target hash address; then, with the help of a preset convolutional neural network model, the initial image features of the initial human body image are determined by combining the target feature parameters and the initial human body image. Subsequently, the initial image features are standardized to finally obtain the target image features; through the local state machine, the number of clusters is determined according to the cleaned image, the image processing instructions and the target image features. At the same time, a target clustering algorithm is selected according to the image processing instructions. Then, the target clustering algorithm, the determined number of clusters, the cleaned image and the target image features are used to perform clustering operations on the initial human body image to obtain the corresponding clustering results. Finally, based on the clustering results, various human image sets are grouped according to image categories. An image allocation and scheduling strategy for each of these human image sets is formulated based on the image processing instructions. This strategy is then used to assign each human image set to a specific image processing component for processing. Finally, the processed image results are sent to the target client via the host computer. In this way, this application can improve the efficiency, accuracy, and convenience of human image processing, and to a certain extent, enhance the user experience.

[0088] The following is combined Figure 3 , Figure 4 , Figure 5 as well as Figure 6 The schematic diagram shown illustrates the technical solution of the embodiments of this application in detail.

[0089] For details, see Figure 3As shown, the user sends an image processing application or related request to the host; the host CPU receives the current request instruction (i.e., image processing instruction) and the image to be processed (i.e., the initial human image), and then sends the instruction and the image to be processed to the dedicated image processor based on clustering technology via the PCIe interface in the form of DMA; the image cache unit of the image clustering preprocessing module of the image processor caches all the image data to be processed, and sends each image data and related processing instructions to the clustering preprocessing unit in sequence; the clustering preprocessing unit of the image clustering preprocessing module of the image processor calculates the grouping of all the images to be processed in sequence according to the input processing instructions and the clustering configuration data set by the user, and completes the grouping of each input image to be processed and marks its corresponding group ID, and then transmits it to the image scheduling unit; the image scheduling unit receives all the clustered preprocessed images. Image data is processed, and the number of images in each group is calculated based on the grouping ID data of each image data according to the processing category grouping. The type and number of computing units are assigned to each group of images according to the image processing complexity of the instruction, and one or more target computing units are determined for the current group, thereby forming a corresponding allocation and scheduling strategy (i.e., image allocation and scheduling strategy). The image scheduling unit sends the input image to be processed and the attention score corresponding to the image to be processed to the target computing unit in sequence according to the allocation and scheduling strategy. The target computing unit performs the corresponding data processing and high-performance computing. After the target computing unit completes the computing task, the computing result is temporarily stored in the consistency cache of the computing unit top-level control board. The computing result is written back to the host in the form of DMA through the PCIe interface. After the CPU performs reasonableness verification, the processing result is fed back to the image processing request end user.

[0090] For the clustering preprocessing unit mentioned above, see [link to relevant documentation]. Figure 4 As shown, the clustering preprocessing unit architecture includes a management information configuration module, an image cleaning module, a feature parameter reading module, a feature extraction module, a feature standardization module, a clustering algorithm application FSM (Finite State Machine) state machine, a clustering algorithm calculation unit (including calculation function units for various clustering algorithms such as K-means, spectral clustering, and DBSCAN clustering), a clustering result evaluation and post-processing module, and an image clustering grouping data output module.

[0091] Specifically, the management information configuration module is responsible for completing the necessary parameterization settings and initialization configurations for other modules, configuring the information preset by the Host terminal; the image module is responsible for receiving user instructions (i.e., image processing instructions) and related raw image data (i.e., initial human images), identifying the category of the raw images, and cleaning the image data according to the user-configured parameters; the feature parameter reading module is used to perform size unification processing and HSV conversion on the input image, and extract the image's feature parameters according to the feature parameter storage address (i.e., the target hash address) in the feature parameter storage table (i.e., the preset feature parameter storage table); the input sequence caching module is used to temporarily store the image cleaning result image data and key information such as image processing instructions; the feature extraction module is responsible for extracting raw image features using a pre-trained deep learning CNN (i.e., Convolutional Neural Network) model based on the feature parameters and image data corresponding to the input image; among them, deep learning CNNs include ResNet (i.e., Residual Network) and VGG (i.e., Visual Geometry). Group (a convolutional neural network model); the feature preprocessing module is responsible for performing feature standardization (i.e., Z-Score) and feature dimensionality reduction (which can be PCA, i.e., principal component analysis) on the input raw image feature data, and performing feature fusion on combinations of multiple features; the clustering algorithm uses an FSM state machine to select an appropriate clustering algorithm and number of clusters based on the feature data encoding and image processing instructions of the input encoded image, thereby controlling the clustering algorithm computing unit to perform clustering calculations on the input image data; the clustering algorithm computing unit deploys K-means (i.e., K-means clustering algorithm), spectral clustering, and DBSCAN (i.e., Density-Based Spatial Clustering of Applications with ... The Noise algorithm (a density-based spatial clustering algorithm) and other clustering algorithms are used to perform clustering calculations as needed. The clustering result evaluation and post-processing module is responsible for evaluating the quality of clustering calculations on the input original image. The image clustering grouping data output module groups the cleaned image data of the original image according to a specific method based on the clustering calculation results. The image clustering grouping data output module is responsible for outputting the unprocessed original image data to the subsequent image scheduling module in groups.

[0092] See Figure 5As shown, the detailed process of the clustering preprocessing unit is as follows: The management information configuration module completes the necessary parameterization settings for the image cleaning module, input buffer module, feature parameter reading module, feature extraction module, feature preprocessing module, clustering algorithm application FSM state machine, clustering result evaluation and post-processing module, clustering algorithm calculation unit, and image clustering grouping data output modules. The image cleaning module receives user instructions and relevant raw image data, identifies the raw image category, and cleans the image data according to the user-configured parameters, i.e., removing damaged and low-quality images, and appropriately adjusting brightness and contrast; then, it sequentially encodes the cleaned image data and image processing instructions and sends them to the input buffer module and feature parameter reading module. The feature parameter reading module sequentially performs size unification processing on the input image, performs color space conversion on the unified size to obtain an HSV image, extracts the Key values ​​(i.e., key point feature descriptors) of the image key points, then performs hash calculation on the Key to obtain the feature parameter storage address of the current image in the feature parameter storage table, and then reads the corresponding feature parameters of the image, that is, separates the HSV image into three independent channels: H, S, and V. Keypoint detectors (such as SIFT and ORB) are used to detect keypoints on the selected channel image. This captures the keypoints, which are then identified as keys. The selected channel is typically V. The parameters and the original image are then transmitted to the feature extraction module. The feature extraction module uses a pre-trained deep learning CNN model to extract features from the original image based on the feature parameters and image data corresponding to the input image, and sends the original image feature data to the feature preprocessing module. The feature normalization module performs feature normalization (Z-Score) and feature reduction (PCA) on the input original image feature data, and performs feature fusion on combinations of multiple features. Finally, it obtains all fused feature data of the current encoded input image and sends it to the FSM state machine used in the clustering algorithm. After receiving the feature data of the input encoded image, the clustering algorithm uses the FSM state machine to begin clustering calculations, which consists of three steps: First, it reads the corresponding image cleaning result image data and image processing instructions from the input buffer module based on the encoding of the input feature data; second, it determines the optimal number of clusters; and third, it determines which clustering algorithm to execute based on the instructions for the current input data. It should be noted that the selection of the target clustering algorithm can be found in Table 1 below, and the flowchart for selecting the clustering algorithm can be shown below. Figure 6As shown, the process involves: first, extracting effective feature parameters and relevant image processing instructions from the image; second, iteratively selecting the optimal clustering algorithm that best matches the current image's feature attributes based on the analytical dimensions of the judgment conditions; and finally, selecting the clustering algorithm for the relevant data labels of the current image, thus completing the algorithm selection. Using the input image encoding as a baseline, the cleaned image data and corresponding input feature data are sent to the target computation component of the clustering algorithm computation unit; the computation results are received from the target computation component of the clustering algorithm computation unit, and the clustering results corresponding to the original image encoding are sent to the clustering result evaluation and post-processing module. The clustering result evaluation and post-processing module evaluates the clustering computation quality of the input original image. If there are any abnormalities, it sends the data to the clustering algorithm application FSM state machine for recalculation. If there are no abnormalities, the clustering computation results and the corresponding cleaned image data of the original image are directly sent to the image clustering grouping data output module. The image clustering grouping data output module, based on the cached clustering calculation results and the corresponding cleaned original image data, groups the cleaned original image data into a thread of 32 pairs (each subsequent single data processing unit processes a maximum of 32 original image data per cycle). This grouping is done according to the clustering calculation results (maximum 32 groups, with each group containing a maximum of 32 cleaned original image data). The image clustering grouping data output module then outputs the original image data, categorized into groups, to the subsequent image scheduling module, which operates in parallel with the clustering preprocessing unit.

[0093] Table 1

[0094] Analysis Dimensions Key Indicators Algorithm selection 1. Feature Dimension Is the dimension greater than 300? For values ​​less than or equal to 300, use K-means; for values ​​greater than 300, use spectral clustering. 2. Sample size Is the sample size greater than 100K? For samples smaller than 100K, spectral clustering is used; otherwise, K-means is used. 3. Proportion of abnormal noise Is the noise anomaly rate greater than 5%? If the proportion is greater than 5%, use DBSCAN clustering; otherwise, use other clustering methods.

[0095] Accordingly, see Figure 7 As shown, this application provides a human image processing device based on clustering technology, applied to an image processor, including:

[0096] Information acquisition module 11 is used to acquire the image processing instructions of the target client sent by the host and the corresponding initial human body image to be processed; the initial human body image is an image including a face and / or human body movements;

[0097] The parameter determination module 12 is used to clean the initial human body image, determine an HSV image based on the obtained cleaned image, and perform a hash operation on the key point feature descriptors determined from the HSV image, so as to determine the corresponding target feature parameters from the preset feature parameter storage table using the obtained target hash address; the preset feature parameter storage table is a table that records the correspondence between the feature parameters of the human body image and the hash address.

[0098] The feature processing module 13 is used to determine the initial image features of the initial human image based on a preset convolutional neural network model and using the target feature parameters and the initial human image, and to standardize the initial image features to obtain the target image features;

[0099] The image clustering module 14 is used to determine the number of clusters based on the cleaned image, the image processing instructions, and the target image features through a local state machine, determine the target clustering algorithm based on the image processing instructions, and cluster the initial human image using the target clustering algorithm, the number of clusters, the cleaned image, and the target image features to obtain the corresponding clustering results.

[0100] Image processing module 15 is used to determine each human body image set grouped by image category based on the clustering results, determine the image allocation and scheduling strategy for each human body image set based on the image processing instructions, allocate each human body image set to each image processing component for image processing using the image allocation and scheduling strategy, and send the obtained image processing results to the target client through the host.

[0101] As can be seen from the above, in this application, the image processing instructions of the target client sent by the host and the corresponding initial human body image to be processed are received first; then, a cleaning operation is performed on the initial human body image, and an HSV image is determined based on the image obtained after cleaning. Then, a hash operation is performed on the key point feature descriptors extracted from the HSV image, and the corresponding target feature parameters are searched and determined from the preset feature parameter storage table based on the obtained target hash address; then, with the help of a preset convolutional neural network model, the initial image features of the initial human body image are determined by combining the target feature parameters and the initial human body image. Subsequently, the initial image features are standardized to finally obtain the target image features; through the local state machine, the number of clusters is determined according to the cleaned image, the image processing instructions and the target image features. At the same time, a target clustering algorithm is selected according to the image processing instructions. Then, the target clustering algorithm, the determined number of clusters, the cleaned image and the target image features are used to perform clustering operations on the initial human body image to obtain the corresponding clustering results. Finally, based on the clustering results, various human image sets are grouped according to image categories. An image allocation and scheduling strategy for each of these human image sets is formulated based on the image processing instructions. This strategy is then used to assign each human image set to a specific image processing component for processing. Finally, the processed image results are sent to the target client via the host computer. In this way, this application can improve the efficiency, accuracy, and convenience of human image processing, and to a certain extent, enhance the user experience.

[0102] In some specific embodiments, the information acquisition module 11 specifically includes:

[0103] The information acquisition unit is used to acquire the image processing instructions and the corresponding initial human body image to be processed sent by the host through a preset interface and based on the direct memory access method from the target client, so that the image processor can save the acquired image processing instructions and the initial human body image to the preset image cache component in the image processor.

[0104] In some specific embodiments, the parameter determination module 12 specifically includes:

[0105] A quality determination unit is used to determine the image quality of the initial human body image;

[0106] The image cleaning unit is used to remove human images that meet preset low quality conditions from the initial human image by using preset image cleaning parameters configured on the target user terminal, and to adjust the brightness and contrast of the initial human image after image removal, so as to complete the image cleaning operation of the initial human image and obtain the corresponding cleaned image.

[0107] In some specific embodiments, the parameter determination module 12 specifically includes:

[0108] The image conversion unit is used to perform size unification processing on the cleaned image and color space conversion on the obtained unified image to obtain an HSV image;

[0109] The address determination unit is used to select any one of the hue channel, saturation channel and brightness channel in the HSV image as the target channel, and use a preset key point detector to determine the key point feature descriptor from the target channel of the HSV image, and perform a hash operation on the key point feature descriptor to obtain the target hash address.

[0110] The parameter determination unit is used to determine the target feature parameters corresponding to the target hash address from a preset feature parameter storage table.

[0111] In some specific embodiments, the feature processing module 13 specifically includes:

[0112] The feature processing unit is used to perform Z-Score normalization and preset feature dimensionality reduction on the initial image features to obtain the corresponding processed image features.

[0113] The feature sending unit is used to perform a preset feature fusion operation on the processed image features to obtain target image features, and then send the target image features to the state machine of the image processor.

[0114] In some specific embodiments, the image clustering module 14 specifically includes:

[0115] An information determination unit is used to determine the feature dimension, feature quantity, and noise anomaly ratio of the target image features based on the image processing instructions.

[0116] The first information comparison unit is used to compare the feature dimension with a preset dimension threshold to obtain a first comparison result.

[0117] The first algorithm determining unit is configured to determine the K-Means clustering algorithm as the target clustering algorithm if the first comparison result indicates that the feature dimension is not greater than the preset dimension threshold.

[0118] The second algorithm determination unit is used to determine the spectral clustering algorithm as the target clustering algorithm if the first comparison result indicates that the feature dimension is greater than the preset dimension threshold.

[0119] The second information comparison unit is used to compare the feature quantity with a preset quantity threshold to obtain a second comparison result;

[0120] The third algorithm determination unit is used to determine the spectral clustering algorithm as the target clustering algorithm if the second comparison result indicates that the feature quantity is less than the preset quantity threshold.

[0121] The fourth algorithm determination unit is used to determine the K-Means clustering algorithm as the target clustering algorithm if the second comparison result indicates that the feature quantity is not less than the preset quantity threshold.

[0122] The third information comparison unit is used to compare the noise anomaly ratio with a preset ratio threshold to obtain a third comparison result.

[0123] The fifth algorithm determination unit is used to determine the DBSCAN clustering algorithm as the target clustering algorithm if the third comparison result indicates that the noise anomaly ratio is greater than the preset ratio threshold.

[0124] The sixth algorithm determination unit is used to determine the spectral clustering algorithm or the K-Means clustering algorithm as the target clustering algorithm if the third comparison result indicates that the proportion of noise anomalies is not greater than the preset proportion threshold.

[0125] In some specific embodiments, the image processing module 15 specifically includes:

[0126] A complexity determination unit is used to determine the image processing complexity based on the image processing instructions;

[0127] The strategy determination unit is used to determine the type and number of image processing components corresponding to each of the human body image sets based on the image processing complexity, and to determine the image allocation and scheduling strategy for each of the human body image sets using the type and the number.

[0128] Furthermore, embodiments of this application also disclose an electronic device, Figure 8 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the human image processing method based on clustering technology disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0129] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0130] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0131] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the human image processing method based on clustering technology disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0132] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned human image processing method based on clustering technology. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0133] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0134] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0135] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0136] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0137] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A human image processing method based on clustering technology, characterized in that, Applied to image processors, including: The system acquires the image processing instructions and corresponding initial human body image to be processed from the target client sent by the host; the initial human body image is an image including a face and / or human body movements. The initial human image is cleaned, an HSV image is determined based on the cleaned image, and a hash operation is performed on the key point feature descriptors determined from the HSV image to determine the corresponding target feature parameters from a preset feature parameter storage table using the obtained target hash address; the preset feature parameter storage table is a table that records the correspondence between the feature parameters of the human image and the hash address. Based on a preset convolutional neural network model, the initial image features of the initial human body image are determined using the target feature parameters and the initial human body image, and the initial image features are standardized to obtain the target image features. The initial human image is clustered using a local state machine, based on the cleaned image, the image processing instructions, and the target image features. The number of clusters is determined, the target clustering algorithm is determined based on the image processing instructions, and the initial human image is clustered using the target clustering algorithm, the number of clusters, the cleaned image, and the target image features to obtain the corresponding clustering results. Based on the clustering results, each human image set is determined by image category. Based on the image processing instructions, an image allocation and scheduling strategy for each human image set is determined. The image allocation and scheduling strategy is used to allocate each human image set to each image processing component for image processing, so that the obtained image processing results are sent to the target client through the host.

2. The human image processing method based on clustering technology according to claim 1, characterized in that, The step of obtaining the image processing instructions for the target client sent by the host and the corresponding initial human image to be processed includes: The image processor acquires the image processing instructions and the corresponding initial human body image to be processed from the target client sent by the host through a preset interface and based on direct memory access, so that the image processor saves the acquired image processing instructions and the initial human body image to a preset image cache component in the image processor.

3. The human image processing method based on clustering technology according to claim 1, characterized in that, The cleaning of the initial human image includes: Determine the image quality of the initial human body image; In the initial human body image, human body images that meet the preset low quality conditions are removed by using preset image cleaning parameters configured on the target user terminal, and the brightness and contrast of the initial human body image after image removal are adjusted to complete the image cleaning operation of the initial human body image and obtain the corresponding cleaned image.

4. The human image processing method based on clustering technology according to claim 1, characterized in that, The process of determining an HSV image based on the obtained cleaned image and performing a hash operation on the key point feature descriptors determined from the HSV image to determine the corresponding target feature parameters from a preset feature parameter storage table using the obtained target hash address includes: The cleaned image is then resized and the resulting resized image is converted to a color space to obtain an HSV image. Select any one of the hue, saturation, and brightness channels in the HSV image as the target channel, and use a preset key point detector to determine key point feature descriptors from the target channel of the HSV image, and perform a hash operation on the key point feature descriptors to obtain the target hash address; The target feature parameters corresponding to the target hash address are determined from the preset feature parameter storage table.

5. The human image processing method based on clustering technology according to claim 1, characterized in that, The standardization process for the initial image features to obtain the target image features includes: The initial image features are subjected to Z-Score normalization and preset feature dimensionality reduction to obtain the corresponding processed image features; The processed image features are subjected to a preset feature fusion operation to obtain target image features, which are then sent to the state machine of the image processor.

6. The human image processing method based on clustering technology according to claim 1, characterized in that, The target clustering algorithm based on the image processing instructions includes: Based on the image processing instructions, determine the feature dimension, feature quantity, and noise anomaly ratio of the target image features; By comparing the feature dimension with a preset dimension threshold, a first comparison result is obtained; If the first comparison result indicates that the feature dimension is not greater than the preset dimension threshold, then the K-Means clustering algorithm is determined as the target clustering algorithm; If the first comparison result indicates that the feature dimension is greater than the preset dimension threshold, then the spectral clustering algorithm is determined as the target clustering algorithm; By comparing the feature quantity with a preset quantity threshold, a second comparison result is obtained; If the second comparison result indicates that the feature quantity is less than the preset quantity threshold, then the spectral clustering algorithm is determined as the target clustering algorithm; If the second comparison result indicates that the feature quantity is not less than the preset quantity threshold, then the K-Means clustering algorithm is determined as the target clustering algorithm; By comparing the noise anomaly ratio with the preset ratio threshold, a third comparison result is obtained; If the third comparison result indicates that the proportion of noise anomalies is greater than the preset proportion threshold, then the DBSCAN clustering algorithm is determined as the target clustering algorithm. If the third comparison result indicates that the proportion of noise anomalies is not greater than the preset proportion threshold, then the spectral clustering algorithm or the K-Means clustering algorithm is determined as the target clustering algorithm.

7. The human image processing method based on clustering technology according to any one of claims 1 to 6, characterized in that, The step of determining the image allocation and scheduling strategy for each of the human image sets based on the image processing instructions includes: The image processing complexity is determined based on the image processing instructions. Based on the image processing complexity, the type and number of image processing components corresponding to each of the human body image sets are determined, and the image allocation and scheduling strategy for each of the human body image sets is determined using the type and the number.

8. A human image processing device based on clustering technology, characterized in that, Applied to image processors, including: The information acquisition module is used to acquire the image processing instructions and the corresponding initial human body image to be processed from the target client sent by the host; the initial human body image is an image including a face and / or human body movements; The parameter determination module is used to clean the initial human body image, determine an HSV image based on the cleaned image, and perform a hash operation on the key point feature descriptors determined from the HSV image, so as to determine the corresponding target feature parameters from the preset feature parameter storage table using the obtained target hash address; the preset feature parameter storage table is a table that records the correspondence between the feature parameters of the human body image and the hash address. The feature processing module is used to determine the initial image features of the initial human image based on a preset convolutional neural network model and using the target feature parameters and the initial human image, and to standardize the initial image features to obtain the target image features; The image clustering module is used to determine the number of clusters based on the cleaned image, the image processing instructions, and the target image features through a local state machine, determine the target clustering algorithm based on the image processing instructions, and cluster the initial human image using the target clustering algorithm, the number of clusters, the cleaned image, and the target image features to obtain the corresponding clustering results. The image processing module is used to determine each human image set grouped by image category based on the clustering results, determine the image allocation and scheduling strategy for each human image set based on the image processing instructions, allocate each human image set to each image processing component for image processing using the image allocation and scheduling strategy, and send the obtained image processing results to the target client through the host.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the human image processing method based on clustering technology as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the human image processing method based on clustering technology as described in any one of claims 1 to 7.

Images