Image transmission method and device based on machine learning, computer equipment and storage medium

Abstract

The invention relates to the field of cloud transmission, and discloses an image transmission method and device based on machine learning, computer equipment and a storage medium, and the method comprises the steps: compressing a to-be-transmitted image through an image compression model matched with a transmission environment, and obtaining a compressed image; and sending the compressed image to a receiving end, so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image through an image optimization model matched with the decompressed image to obtain a target image. According to the method, the most suitable image compression model is dynamically selected according to the transmission environment, and it is ensured that image compression and transmission can be efficiently carried out under different network conditions and equipment computing capabilities; a receiving end adopts an image optimization model to recover details, and the image quality is improved. The technology can be applied to high-precision face recognition and identity verification in a financial scene, supports high-quality remote transmission and auxiliary diagnosis of medical images in a medical scene, and effectively guarantees image definition and transmission efficiency in key services.

Description

Technical Field

[0001] The present invention relates to the field of cloud transmission, and in particular to an image transmission method, device, computer equipment and storage medium based on machine learning. Background Art

[0002] With the rise of the smart city concept, image transmission technology is becoming increasingly important as the foundation for key applications such as intelligent surveillance, traffic management, financial services, and medical diagnosis. In financial scenarios, high-definition, real-time image transmission is crucial for remote identity verification and over-the-counter transactions. In the medical field, the remote transmission of high-quality medical images is directly related to the accuracy of remote diagnosis and surgical guidance. However, in practice, network image transmission faces many challenges, such as bandwidth limitations that hinder the rapid transmission of high-resolution images and videos, and image quality degradation caused by compression processing to accommodate limited bandwidth, including blurring and color deviation. Especially in real-time transmission scenarios, such as video conferencing or live streaming, network latency not only reduces interaction efficiency but also compromises the realism of the user experience. Furthermore, packet loss in unstable network environments, which causes incomplete image information and pixelation, further exacerbates these issues.

[0003] In view of this, it is necessary to find a new image transmission method that can reduce the bit rate while ensuring image quality. Summary of the Invention

[0004] Embodiments of the present invention provide an image transmission method, apparatus, computer device, and storage medium based on machine learning to solve the problem of reducing the bit rate while ensuring image quality.

[0005] A machine learning-based image transmission method, comprising: Acquire the transmission environment of the image to be transmitted; Acquire an image compression model that matches the transmission environment; compressing the image to be transmitted using the image compression model to obtain a compressed image; The compressed image is sent to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and the decompressed image is optimized by an image optimization model matching the decompressed image to obtain a target image.

[0006] Optionally, obtaining a transmission environment of the image to be transmitted includes: Collect the network status of the sender; Obtaining computing capability information of the sending end; The transmission environment is determined according to the network status and the computing capability information.

[0007] Optionally, acquiring an image compression model matching the transmission environment includes: Acquire a first model parameter set matching the network state; Acquire a second model parameter set that matches the computing capability information; Calculating the intersection of the first model parameter set and the second model parameter set to obtain a feasible parameter subset; A compression model is set according to the feasible parameter subset to obtain the image compression model.

[0008] Optionally, compressing the image to be transmitted by using the image compression model to obtain a compressed image includes: Extracting texture features of the image to be transmitted; Clustering is performed according to the texture features to divide the image to be transmitted into a plurality of image regions; Adaptively compressing each of the image regions based on the texture features to obtain a regional compressed image; All the regional compressed images are integrated and optimized to obtain the compressed image.

[0009] Optionally, sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model matching the decompressed image to obtain a target image includes: At the receiving end, obtaining compression parameters of the compressed image; Acquire the image optimization model that matches the compression parameters.

[0010] Optionally, sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model matching the decompressed image to obtain a target image includes: At the receiving end, preprocessing the decompressed image to obtain a preprocessed image; Processing the preprocessed image through a generative adversarial network to obtain a detailed reconstructed image; Adaptively enhancing and optimizing edges of the detail reconstructed image to obtain an enhanced image; Post-processing optimization is performed on the enhanced image to obtain the target image.

[0011] Optionally, before sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model matching the decompressed image to obtain a target image, the method further includes: Obtain historical compressed images and historical target images that match the specified compression parameter range; Inputting the historical compressed image and the historical target image into an initial image optimization model for training; The trained initial image optimization model is determined as the image optimization model.

[0012] An image transmission device based on machine learning, comprising: A transmission environment acquisition module is used to acquire the transmission environment of the image to be transmitted; A compression model acquisition module is used to acquire an image compression model that matches the transmission environment; An image compression module, configured to compress the image to be transmitted using the image compression model to obtain a compressed image; The image decompression optimization module is used to send the compressed image to the receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizes the decompressed image through an image optimization model matching the decompressed image to obtain a target image.

[0013] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned machine learning-based image transmission method when executing the computer program.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned image transmission method based on machine learning.

[0015] The above-mentioned machine learning-based image transmission method, device, computer equipment and storage medium dynamically select the most suitable image compression model according to the transmission environment, ensuring efficient image compression and transmission under different network conditions and device computing power; this adaptive compression strategy not only ensures high-definition image transmission, but also significantly reduces the bandwidth required for transmission, achieving low-bitrate transmission; the receiving end uses the image optimization model to further process the decompressed image, effectively restores image details and improves overall quality, and ultimately obtains a high-quality target image. In financial scenarios, the present invention can improve the security and efficiency of transactions while ensuring the accuracy of identity authentication, especially in remote financial services and real-time face recognition applications. In the medical field, the present invention enables high-definition medical images to be transmitted quickly and accurately in bandwidth-constrained environments, greatly supporting key medical services such as remote diagnosis and treatment and surgical guidance, and improving the accuracy and timeliness of diagnosis. The present invention improves image transmission efficiency and visual effects as a whole, and is particularly suitable for application scenarios where high-definition images need to be transmitted in bandwidth-constrained environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0017] Figure 1 is a flow chart of an image transmission method based on machine learning in one embodiment of the present invention; Figure 2 is a schematic diagram of an image transmission device based on machine learning in one embodiment of the present invention; Figure 3 FIG. 1 is a schematic diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] In one embodiment, if Figure 1 As shown, a method for image transmission based on machine learning is provided, comprising the following steps S10 to S40.

[0020] S10: Acquire the transmission environment of the image to be transmitted.

[0021] Understandably, the image to be transmitted can be a picture or video. The transmission environment refers to the dynamic parameters involved in the image transmission process, such as network bandwidth, latency, packet loss rate, and device performance (including the sender and receiver). For example, in a medical image transmission scenario, the real-time network bandwidth is detected through TCP congestion status (e.g., 5G network fluctuates between 0.05-0.5Gbps), and the performance rating of the sender (e.g., high, medium, and low) is also obtained.

[0022] S20: Acquire an image compression model that matches the transmission environment.

[0023] Understandably, image compression models can be dynamically selected based on the transmission environment, balancing image quality and transmission efficiency. For example, in low-bandwidth environments, JPEG lossy compression (quantization factor 60) combined with Base64 encoding can be used to compress a 1080P image from 2MB to 50KB. In high-latency environments (such as satellite communications), the WebP format combined with motion-compensated compression can be used to reduce the keyframe interval to improve real-time performance. In edge computing scenarios, a lightweight encoding model can be deployed, reducing the model size by 67% through 8-bit quantization.

[0024] S30: compress the image to be transmitted using the image compression model to obtain a compressed image.

[0025] Understandably, the image compression model can be a deep learning model. The image compression model can be used to perform in-depth analysis of the image to be transmitted, dynamically adjust the compression ratio according to the texture characteristics of the image content, and generate a compressed image. For example, in the texture-rich area of ​​the image to be transmitted, the image compression model will choose a higher fidelity for compression to ensure that the image details are preserved; while for the background or smooth areas, a higher compression rate can be used to reduce the amount of data and improve transmission efficiency. In a specific application example, for CT images, the image compression model first decomposes the image into high-frequency (details) and low-frequency (contour) sub-bands through wavelet transform, and then adaptively quantizes the high-frequency part, while retaining lossless compression of the low-frequency part. Through the adaptive compression of the image compression model, not only the quality of the compressed image is improved, but also the storage space and bandwidth consumption can be effectively reduced, which is particularly suitable for network environments with limited bandwidth.

[0026] S40: Send the compressed image to a receiving end, so that the receiving end decompresses the compressed image into a decompressed image, and optimizes the decompressed image using an image optimization model that matches the decompressed image to obtain a target image.

[0027] Understandably, the compressed image can be sent to the receiving end using the video transmission protocol of the current transmission environment, such as HTTP multipart transmission (Multipart / JPEG). The receiving end can be a device terminal such as a laptop, mobile phone, or server. The receiving end first decompresses the compressed image to a decompressed image, and then optimizes the decompressed image using a machine learning-based image optimization model to obtain the target image. The image optimization model is a deep learning-based post-processing model used to restore details lost in compression (such as super-resolution, denoising, and texture generation).

[0028] This embodiment dynamically selects the most suitable image compression model based on the transmission environment, ensuring efficient image compression and transmission under different network conditions and device computing capabilities; this adaptive compression strategy not only ensures high-definition image quality during transmission, but also significantly reduces the bandwidth required for transmission, achieving low-bitrate transmission; the receiving end uses the image optimization model to further process the decompressed image, effectively restoring image details and improving overall quality, ultimately obtaining a high-quality target image. In financial scenarios, this embodiment can improve the security and efficiency of transactions while ensuring the accuracy of identity authentication, especially in remote financial services and real-time face recognition applications. In the medical field, this embodiment enables high-definition medical images to be transmitted quickly and accurately in bandwidth-constrained environments, greatly supporting key medical services such as remote diagnosis and treatment, surgical guidance, and improving the accuracy and timeliness of diagnosis. This embodiment improves image transmission efficiency and visual effects as a whole, and is particularly suitable for application scenarios where high-definition images need to be transmitted in bandwidth-constrained environments.

[0029] Optionally, step S10, i.e., obtaining a transmission environment of the image to be transmitted, includes: S101, collecting the network status of the sending end; S102: Acquire computing capability information of the sending end; S103: Determine the transmission environment according to the network status and the computing capability information.

[0030] The sender's network status refers to parameters such as the real-time available network bandwidth, latency, packet loss rate, and signal stability during image transmission. This network status directly impacts transmission quality and the upper bitrate limit. Speed ​​test tools can be used to measure the current available bandwidth in real time. Ping tests can be used to obtain end-to-end latency. Wireless network card signal strength can be read (e.g., RSSI = -70dBm corresponds to medium signal quality). The sender's computing power refers to the sender's hardware resources (e.g., CPU / GPU computing power, memory capacity, storage speed) and the encoding acceleration capabilities supported by the software framework.

[0031] After obtaining the network status and computing power information of the sender, the transmission environment can be determined by combining the network status and computing power information. In some examples, the transmission environment can be divided into a weak network + high computing power environment, a weak network + low computing power environment, a strong network + high computing power environment, and a strong network + low computing power environment.

[0032] This embodiment accurately assesses the current transmission environment by collecting network status and computing power information from the transmitter, laying the foundation for the subsequent selection of the most suitable image compression model and ensuring efficient preparation and optimization of image data under different network conditions and hardware limitations. After determining the transmission environment, image compression parameters and algorithms can be adjusted specifically to achieve high-definition, low-bitrate transmission while ensuring image quality and transmission efficiency, thereby providing a high-quality image transmission experience under limited resources.

[0033] Optionally, step S20, i.e., obtaining an image compression model matching the transmission environment, includes: S201. Acquire a first model parameter set matching the network state; S202: Acquire a second model parameter set that matches the computing capability information; S203: Calculate the intersection of the first model parameter set and the second model parameter set to obtain a feasible parameter subset; S204: Setting a compression model according to the feasible parameter subset to obtain the image compression model.

[0034] Understandably, the first set of model parameters (such as quantization step size, entropy coding mode, and ROI region division threshold) that adapt to the current channel conditions can be screened based on network status indicators such as real-time network bandwidth, latency, and packet loss rate. For example, in a 5G network bandwidth fluctuation scenario (bandwidth 50-100Mbps, latency <50ms), based on the web page framework, the following semantic decoupling parameters are selected: global reference image quantization parameter QP = 35 (compressed to 0.05bpp); semantic mask encoding uses CABAC entropy coding (context-based adaptive binary arithmetic coding).

[0035] Based on device CPU / GPU computing power, memory capacity, and other computing capability indicators, you can select a set of second model parameters (such as model complexity, number of parallel threads, and hardware acceleration interfaces) that can be completed within a limited time. For example, in edge computing device scenarios, enable the ORAG model attention guidance mechanism (object-relational attention guidance mechanism) and set the number of cross-attention layers to 8; based on the DCT quantization strategy (discrete cosine transform quantization strategy), select the GPU-accelerated 8×8 block processing mode.

[0036] A logical AND operation is performed on the first and second model parameter sets to select a feasible parameter subset that satisfies both channel conditions and device performance. A compression model is then set based on this feasible parameter subset, dynamically binding the algorithm module with hardware resources to form an image compression model.

[0037] This embodiment obtains a first set of model parameters that matches the network state and a second set of model parameters that matches the computing power, and calculates their intersection. This allows for the selection of an optimal compression model parameter combination that satisfies both the current network conditions and the device's computing power. This effectively enables dynamic adaptation of the compression model, ensuring efficient compression performance in varying transmission environments. The image compression model, based on a subset of feasible parameters, not only improves resource utilization but also further ensures high-quality transmission of high-definition images at low bit rates.

[0038] Optionally, step S30, i.e., compressing the image to be transmitted using the image compression model to obtain a compressed image, includes: S301, extracting texture features of the image to be transmitted; S302, performing clustering according to the texture features, and dividing the image to be transmitted into a plurality of image regions; S303, adaptively compressing each of the image regions based on the texture features to obtain a regional compressed image; S304: Integrate and optimize all the regional compressed images to obtain the compressed image.

[0039] Understandably, texture features of the image to be transmitted can be extracted. Texture features are quantitative indicators that reflect the spatial distribution of local grayscale in an image. They are used to describe the frequency, directionality, and repeatability of brightness changes within a pixel neighborhood. In some examples, texture features of the image to be transmitted can be extracted using a gray-level co-occurrence matrix (GLCM) or a local binary pattern (LBP).

[0040] Based on texture similarity, images can be segmented into regions with homogeneous texture characteristics, enabling the aggregation of semantically related regions. For this purpose, algorithms such as the improved DBSCAN density clustering algorithm (an improved density-based arbitrary shape clustering algorithm) and K-means dynamic clustering can be used.

[0041] Quantization parameters, transform modes, and entropy coding strategies can be dynamically adjusted based on the texture complexity of different regions, achieving adaptive compression and generating regional compressed images. For example, hierarchical compression is implemented in smart city surveillance systems: for smooth areas (such as the sky), the JPEG standard quantization table is used, with a QP (quantization parameter) of 40, a DCT block size (discrete cosine transform block size) of 8×8, a bitrate of 0.05bpp, and a PSNR (peak signal-to-noise ratio) maintained above 40dB. For areas with complex textures (such as traffic intersections), the HEVC (High Efficiency Video Coding) RQT (Residual Quadtree Transform) is enabled, with the CU size (coding unit) adaptively ranging from 32×32 to 8×8. Directional DCT is used for vehicle outlines to preserve edge information. For special areas (such as text on electronic screens), semantic compression with Gabor filter preprocessing is applied, achieving a 22% bitrate reduction compared to traditional methods at the same PSNR.

[0042] The differentially compressed regional image can be subjected to boundary fusion, bitstream reassembly, and visual quality enhancement to ultimately obtain a compressed image. The Dual-Tree Complex Wavelet Transform (DT-CWT) can be used to eliminate blocking artifacts and achieve boundary fusion. Based on ROI (Region of Interest) priority packaging, facial regions can be assigned a 3x bitrate weight to achieve bitstream reassembly. SRGAN (Super-Resolution Generative Adversarial Network) super-resolution reconstruction of the compressed cloud region improves the SSIM (Structural Similarity Index) by 0.15.

[0043] This embodiment extracts image texture features and combines them with cluster analysis to divide the image into multiple regions with similar characteristics, achieving adaptive compression processing for different regions. It dynamically adjusts the compression strategy based on regional complexity, effectively reducing the overall bitrate while preserving important visual information. After regional compression, it integrates and optimizes the compressed regions, further improving the quality consistency and transmission efficiency of the compressed images. This embodiment overall improves the transmission performance of high-definition images under limited bandwidth conditions, achieving the optimization goal of high quality and low bitrate.

[0044] Optionally, step S40, i.e., sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model matching the decompressed image to obtain a target image, includes: S401. At the receiving end, obtaining compression parameters of the compressed image; S402: Acquire the image optimization model that matches the compression parameters.

[0045] Understandably, compression parameters refer to the algorithm configuration or quantization standards used during the compression process. These include core parameters such as compression quality level, quantization matrix, and coding mode, which directly impact image reconstruction quality and data volume. The receiver can extract these parameters from the compressed image. For example, the receiver can parse the quantization table in the JPEG file header to obtain the quantization step size after the DCT transform. Another example is analyzing the coding unit division method (e.g., 8x8 or 16x16 blocks) in the video stream to determine whether inter-frame motion compensation is used, thereby identifying the coding mode.

[0046] An image optimization model can be dynamically selected based on compression parameters to repair compression losses (such as blur and artifacts) or enhance image detail. For example, if a quantization step size > 50 is detected, a GAN (Generative Adversarial Network)-based super-resolution model (such as the Enhanced Super-Resolution Generative Adversarial Network (ESRGAN)) is selected to restore texture details through adversarial training. Alternatively, if compression uses 8x8 block DCT, a U-Net (U-Net) restoration model is used to locally remove blocking artifacts.

[0047] This embodiment obtains the compression parameters of the compressed image at the receiving end and matches the corresponding image optimization model accordingly, which can achieve targeted repair and enhancement of the decompressed image; ensures that the optimal optimization strategy can be selected under different compression conditions, effectively restores image details and improves visual quality; and through the parameter-driven model matching mechanism, improves the intelligence level and adaptability of image processing.

[0048] Optionally, step S40, i.e., sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model matching the decompressed image to obtain a target image, includes: S403: Preprocessing the decompressed image at the receiving end to obtain a preprocessed image; S404, processing the preprocessed image through a generative adversarial network to obtain a detailed reconstructed image; S405, performing adaptive enhancement and edge optimization on the detail reconstructed image to obtain an enhanced image; S406: Perform post-processing optimization on the enhanced image to obtain the target image.

[0049] Understandably, at the receiving end, the decompressed image is subjected to preprocessing operations such as noise suppression, contrast enhancement, and geometric correction to obtain a preprocessed image. By preprocessing the decompressed image, high-quality input is provided for subsequent processing.

[0050] Generative adversarial networks (GANs) are adversarial learning frameworks consisting of a generator (which simulates image reconstruction) and a discriminator (which assesses image authenticity). By processing pre-processed images through GANs, high-frequency details lost in compression are restored, resulting in detailed reconstructed images.

[0051] Adaptive enhancement dynamically adjusts enhancement parameters based on image content, such as adaptive gamma correction and deep learning enhancement. Edge optimization combines multi-scale edge detection to optimize outline clarity, using, for example, a modified Canny operator: σ = 1.2 to detect vehicle outlines, and σ = 2.0 to extract building edges. Adaptive enhancement and edge optimization are performed on the detailed reconstructed image to produce an enhanced image.

[0052] Post-processing optimization includes, but is not limited to, color correction, sharpening filtering, and anti-aliasing. This optimization can further enhance the visual quality of the target image. For example, improved Laplacian pyramid sharpening can be used to enhance the movement paths of athletes in a football game scene.

[0053] This embodiment effectively restores image detail by preprocessing and generative adversarial network processing the decompressed image at the receiving end, significantly improving the clarity and realism of the reconstructed image. Adaptive enhancement and edge optimization further improve image quality, making image details sharper and more natural. Post-processing optimization enhances the overall visual consistency and comfort of the image. This embodiment achieves high-quality image restoration under low bitrate transmission conditions, significantly enhancing the user's visual experience.

[0054] Optionally, before step S40, that is, before sending the compressed image to the receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model matching the decompressed image to obtain a target image, the method further includes: S41, acquiring a historical compressed image and a historical target image that match a specified compression parameter range; S42, inputting the historical compressed image and the historical target image into an initial image optimization model for training; S43: Determine the trained initial image optimization model as the image optimization model.

[0055] It is understood that multiple compression parameter ranges can be set, with the specified compression parameter range being one of the multiple compression parameter ranges. The compression parameters of the compressed image are within the specified compression parameter range. Historical compressed images and historical target images matching the specified compression parameter range can be obtained and used to train an initial image optimization model, ultimately obtaining an image optimization model. Different image optimization models can be trained for different compression parameter ranges, further improving the optimization quality of the image optimization model.

[0056] This embodiment constructs a targeted training dataset by acquiring historical compressed images and corresponding historical target images that match a specified compression parameter range. This allows the model to learn the laws governing image restoration under specific compression distortions. This data is then fed into the initial image optimization model for training, improving the model's ability to recover image degradation in real-world transmission scenarios. The trained image optimization model possesses enhanced generalization and adaptability, enabling it to more accurately restore high-definition image details. This embodiment effectively enhances the quality and stability of image reconstruction at the receiving end, further optimizing the overall effectiveness of high-definition, low-bitrate transmission.

[0057] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0058] In one embodiment, a machine learning-based image transmission device is provided, which corresponds one-to-one to the machine learning-based image transmission method in the above embodiment. Figure 2 As shown, the image transmission device based on machine learning includes: A transmission environment acquisition module 10 is used to acquire the transmission environment of the image to be transmitted; A compression model acquisition module 20 is used to acquire an image compression model that matches the transmission environment; An image compression module 30 is configured to compress the image to be transmitted using the image compression model to obtain a compressed image; The image decompression optimization module 40 is used to send the compressed image to the receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizes the decompressed image through an image optimization model matching the decompressed image to obtain a target image.

[0059] Optionally, the transmission environment acquisition module 10 includes: A network status collecting unit, used for collecting the network status of the sending end; A computing capability information acquisition unit, configured to acquire computing capability information of the sending end; The transmission environment determining unit is configured to determine the transmission environment according to the network status and the computing capability information.

[0060] Optionally, the compression model acquisition module 20 includes: A first parameter set unit, configured to obtain a first model parameter set matching the network state; A second parameter set unit, configured to obtain a second model parameter set matching the computing capability information; A feasible parameter subset unit, configured to calculate an intersection of the first model parameter set and the second model parameter set to obtain a feasible parameter subset; The compression model setting unit is used to set the compression model according to the feasible parameter subset to obtain the image compression model.

[0061] Optionally, the image compression module 30 includes: A texture extraction unit, used for extracting texture features of the image to be transmitted; a clustering unit, configured to perform clustering according to the texture features and divide the image to be transmitted into a plurality of image regions; A region compression unit, configured to adaptively compress each of the image regions based on the texture features to obtain a region compressed image; The compressed image generating unit is used to integrate and optimize all the regional compressed images to obtain the compressed image.

[0062] Optionally, the image decompression optimization module 40 includes: A compression parameter obtaining unit, configured to obtain, at the receiving end, the compression parameters of the compressed image; The matching optimization model unit is used to obtain the image optimization model that matches the compression parameters.

[0063] Optionally, the image decompression optimization module 40 includes: A preprocessing unit, configured to preprocess the decompressed image at the receiving end to obtain a preprocessed image; a detail reconstruction unit, configured to process the preprocessed image through a generative adversarial network to obtain a detail reconstructed image; An image enhancement unit, configured to adaptively enhance and optimize edges of the detail reconstructed image to obtain an enhanced image; A post-processing optimization unit is used to perform post-processing optimization on the enhanced image to obtain the target image.

[0064] Optionally, the image transmission device based on machine learning further includes an optimization model construction module, wherein the optimization model construction module is used to: Obtain historical compressed images and historical target images that match the specified compression parameter range; Inputting the historical compressed image and the historical target image into an initial image optimization model for training; The trained initial image optimization model is determined as the image optimization model.

[0065] For the specific definition of the image transmission device based on machine learning, please refer to the definition of the image transmission method based on machine learning above, which will not be repeated here. The various modules in the above-mentioned image transmission device based on machine learning can be implemented in whole or in part by software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0066] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data involved in the image transmission method based on machine learning. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements an image transmission method based on machine learning.

[0067] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable by the processor. When the processor executes the computer program, the machine learning-based image transmission method described in the above embodiment is implemented. To avoid repetition, this description is omitted. Alternatively, when the processor executes the computer program, the functions of the various modules / units described in the machine learning-based image transmission apparatus embodiment are implemented. To avoid repetition, this description is omitted.

[0068] In one embodiment, a computer-readable storage medium is provided, which stores a computer program. When executed by a processor, the computer program implements the machine learning-based image transmission method described in the above embodiment. To avoid repetition, this description is omitted. Alternatively, when executed by a processor, the computer program implements the functions of the various modules / units in the machine learning-based image transmission device described in the above embodiment. To avoid repetition, this description is omitted.

[0069] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0070] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0071] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A machine learning-based image transmission method, characterized in that: include: Acquire the transmission environment of the image to be transmitted; Acquire an image compression model that matches the transmission environment; compressing the image to be transmitted using the image compression model to obtain a compressed image; The compressed image is sent to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and the decompressed image is optimized by an image optimization model matching the decompressed image to obtain a target image.

2. The image transmission method based on machine learning according to claim 1, characterized in that The step of obtaining the transmission environment of the image to be transmitted includes: Collect the network status of the sender; Obtaining computing capability information of the sending end; The transmission environment is determined according to the network status and the computing capability information.

3. The image transmission method based on machine learning according to claim 2, characterized in that The acquiring of the image compression model matching the transmission environment includes: Acquire a first model parameter set matching the network state; Acquire a second model parameter set that matches the computing capability information; Calculating the intersection of the first model parameter set and the second model parameter set to obtain a feasible parameter subset; A compression model is set according to the feasible parameter subset to obtain the image compression model.

4. The image transmission method based on machine learning according to claim 1, characterized in that The step of compressing the image to be transmitted by using the image compression model to obtain a compressed image includes: Extracting texture features of the image to be transmitted; Clustering is performed according to the texture features to divide the image to be transmitted into a plurality of image regions; Adaptively compressing each of the image regions based on the texture features to obtain a regional compressed image; All the regional compressed images are integrated and optimized to obtain the compressed image.

5. The image transmission method based on machine learning according to claim 1, characterized in that The step of sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model that matches the decompressed image to obtain a target image includes: At the receiving end, obtaining compression parameters of the compressed image; Acquire the image optimization model that matches the compression parameters.

6. The image transmission method based on machine learning according to claim 1, characterized in that: The step of sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model that matches the decompressed image to obtain a target image includes: At the receiving end, preprocessing the decompressed image to obtain a preprocessed image; Processing the preprocessed image through a generative adversarial network to obtain a detailed reconstructed image; Adaptively enhancing and optimizing edges of the detail reconstructed image to obtain an enhanced image; Post-processing and optimization are performed on the enhanced image to obtain the target image.

7. The image transmission method based on machine learning according to claim 1, characterized in that: Before sending the compressed image to a receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizing the decompressed image using an image optimization model matching the decompressed image to obtain a target image, the method further includes: Obtain historical compressed images and historical target images that match the specified compression parameter range; Inputting the historical compressed image and the historical target image into an initial image optimization model for training; The trained initial image optimization model is determined as the image optimization model.

8. An image transmission device based on machine learning, characterized in that: include: A transmission environment acquisition module is used to acquire the transmission environment of the image to be transmitted; A compression model acquisition module is used to acquire an image compression model that matches the transmission environment; An image compression module, configured to compress the image to be transmitted using the image compression model to obtain a compressed image; The image decompression optimization module is used to send the compressed image to the receiving end so that the receiving end decompresses the compressed image into a decompressed image, and optimizes the decompressed image through an image optimization model matching the decompressed image to obtain a target image.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the image transmission method based on machine learning according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the image transmission method based on machine learning according to any one of claims 1 to 7 is implemented.

Images