Image quality adjustment method and system based on adaptive network bandwidth

Abstract

The present invention discloses an image quality adjustment method and system for adaptive network bandwidth, which relates to the field of image transmission technology, including: predicting network bandwidth and determining predicted transmission bandwidth; generating image quality adjustment instructions for target transmission images by discriminating bandwidth compression ratio; performing adversarial network training with low-order and high-order resolution to obtain high-order reconstructors; activating high-order reconstructors based on image quality adjustment instructions, performing high-order reconstruction and loss compensation on target transmission images, and determining adjusted transmission images, wherein loss compensation includes low-light compensation and content loss compensation; and visualizing the adjusted transmission images on a sink display interface. The present invention solves the technical problems of the prior art in that image transmission quality decreases and transmission efficiency is low under network bandwidth fluctuations, and achieves the technical effect of realizing high-quality image transmission and improving transmission efficiency under unstable network bandwidth conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of image transmission, and in particular to an image quality adjustment method and system for adaptive network bandwidth. Background Art

[0002] With the popularization of Internet applications and the increasing demand for high-definition image transmission, the complexity and instability of the network environment have put forward higher requirements for image transmission quality. In practical applications, due to the volatility of network bandwidth, the image transmission process often suffers from transmission quality degradation, which manifests as image blur, distortion, reduced resolution, etc. This situation is particularly obvious in a network environment with low bandwidth or frequent bandwidth fluctuations, which seriously affects the stability of image transmission. Summary of the invention

[0003] The present application provides an image quality adjustment method and system for adaptive network bandwidth, which are used to solve the technical problems of the prior art in that image transmission quality is reduced and transmission efficiency is low under network bandwidth fluctuations.

[0004] In view of the above problems, the present application provides a method and system for adjusting image quality based on adaptive network bandwidth.

[0005] In a first aspect of the present application, a method for adjusting image quality by adaptive network bandwidth is provided, the method comprising:

[0006] The network bandwidth is predicted by using a time series model to determine the transmission prediction bandwidth, wherein the transmission prediction bandwidth is identified by a bandwidth compression ratio, wherein a target transmission image from a source to a destination is used as an image transmission task; for the target transmission image, by discriminating the bandwidth compression ratio, a picture quality adjustment instruction is generated, wherein a preset compression ratio is used as a judgment criterion; adversarial network training is performed based on the low-order-high-order resolution to obtain a high-order reconstructor, wherein the time resolution, spatial resolution and color resolution are used as adjustment targets, and the high-order reconstructor is a generation part of a generative-discriminative architecture retained after training; based on the picture quality adjustment instruction, the high-order reconstructor is activated, and the target transmission image is subjected to high-order reconstruction and loss compensation to determine an adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation; and the adjusted transmission image is visualized on a destination display interface.

[0007] A second aspect of the present application provides an image quality adjustment system for adaptive network bandwidth, the system comprising:

[0008] A network bandwidth prediction module, wherein the network bandwidth prediction module uses a time series model to perform network bandwidth prediction and determine a transmission prediction bandwidth, wherein the transmission prediction bandwidth is marked with a bandwidth compression ratio, wherein a target transmission image from a source to a destination is an image transmission task; a picture quality adjustment instruction generation module, wherein the picture quality adjustment instruction generation module generates a picture quality adjustment instruction for a target transmission image by discriminating the bandwidth compression ratio, wherein a preset compression ratio is used as a judgment criterion; a high-order reconstructor acquisition module, wherein the high-order reconstructor acquisition module performs adversarial network training with a low-order-high-order resolution to obtain a high-order reconstructor, wherein time resolution, spatial resolution and color resolution are adjustment targets, and the high-order reconstructor is a generation part of a generative-discriminative architecture retained after training; an adjusted transmission image determination module, wherein the adjusted transmission image determination module activates a high-order reconstructor based on the picture quality adjustment instruction, performs high-order reconstruction and loss compensation on the target transmission image, and determines an adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation; and a visualization module, wherein the visualization module visualizes the adjusted transmission image on a destination display interface.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] The present application uses a time series model to predict network bandwidth and determine a predicted transmission bandwidth, wherein the predicted transmission bandwidth is identified by a bandwidth compression ratio, wherein a target transmission image from a source to a destination is an image transmission task; for the target transmission image, by discriminating the bandwidth compression ratio, a picture quality adjustment instruction is generated, wherein a preset compression ratio is used as a judgment criterion; adversarial network training is performed with a low-order-high-order resolution to obtain a high-order reconstructor, wherein time resolution, spatial resolution and color resolution are used as adjustment targets, and the high-order reconstructor is a generation part of a generation-discrimination architecture retained after training; based on the picture quality adjustment instruction, a high-order reconstructor is activated, and a high-order reconstruction and loss compensation are performed on the target transmission image to determine an adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation; and the adjusted transmission image is visualized on a destination display interface. The present invention solves the technical problems in the prior art of reduced image transmission quality and low transmission efficiency under network bandwidth fluctuations. The network bandwidth is predicted by a time series model, image quality adjustment instructions are generated, and an adversarial network is used to obtain a high-order reconstructor. The transmitted image is reconstructed in a high-order manner and compensated for low light and content loss. Finally, the adjusted image is output as a visualization, thereby achieving the technical effect of realizing high-quality image transmission and improving transmission efficiency under conditions of unstable network bandwidth. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0012] Figure 1 A schematic diagram of a flow chart of an image quality adjustment method for adaptive network bandwidth provided in an embodiment of the present application;

[0013] Figure 2 A schematic diagram of the structure of an image quality adjustment system for adaptive network bandwidth provided in an embodiment of the present application.

[0014] Description of the accompanying drawings: network bandwidth prediction module 11, image quality adjustment instruction generation module 12, high-order reconstruction device acquisition module 13, adjustment transmission image determination module 14, visualization module 15. DETAILED DESCRIPTION

[0015] The present application provides an image quality adjustment method and system for adaptive network bandwidth, aiming to solve the technical problems of the prior art in that the image transmission quality is reduced and the transmission efficiency is low when the network bandwidth fluctuates. The network bandwidth is predicted by a time series model, image quality adjustment instructions are generated, and a high-order reconstructor is obtained by using an adversarial network. The transmitted image is reconstructed in a high-order manner and compensated for low light and content loss, and finally the adjusted image is output visually, thereby achieving the technical effect of realizing high-quality image transmission and improving transmission efficiency under the condition of unstable network bandwidth.

[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0017] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0018] Embodiment 1, as Figure 1 As shown, the present application provides an image quality adjustment method for adaptive network bandwidth, the method comprising:

[0019] Step S100: using a time series model to predict network bandwidth and determine a transmission prediction bandwidth, wherein the transmission prediction bandwidth is identified by a bandwidth compression ratio, wherein a target transmission image from a source to a destination is an image transmission task.

[0020] In the embodiment of the present application, the target transmission image from the source to the destination is first used as the core data of the image transmission task, and its transmission path and network environment are clarified; the historical bandwidth data and real-time traffic information of the transmission path are analyzed through a time series model, and the available bandwidth status in a specific time period in the future is predicted to determine the transmission prediction bandwidth. The transmission prediction bandwidth reflects the actual available bandwidth level of the network during the transmission process, and is also marked with a bandwidth compression ratio, which indicates the degree of compression of the image transmission quality by the current network bandwidth compared to normal bandwidth conditions.

[0021] Furthermore, in the method provided in the embodiment of the application, the determining of the transmission prediction bandwidth further includes:

[0022] For the target transmission image, a transmission path from a source to a destination is obtained, wherein the transmission path is marked with a transmission time zone; using the time series model, a network bandwidth prediction is performed on the transmission path to determine the transmission prediction bandwidth; based on the transmission prediction bandwidth, the bandwidth compression ratio is calculated.

[0023] In the embodiment of the present application, for the target transmission image, the transmission path from the source to the destination is obtained. Specifically, through the traceroute tool, each intermediate node from the source to the destination is gradually detected to obtain the network connection information in the path, including node address, response time and other data, so as to clarify the complete transmission path. In this process, based on the geographical information or IP location of the node, the transmission time zone passed by the path is identified.

[0024] Then, the time series model is used to predict the network bandwidth. Specifically, the historical bandwidth data of the nodes in the transmission path is obtained through the network traffic monitoring system as the input of the time series model. The ARIMA model is selected to analyze the time correlation and trend reflected in the historical bandwidth data, and the bandwidth status of the transmission path in a specific time period in the future is predicted through the model, so as to determine the transmission prediction bandwidth, which is used to describe the future network resource availability.

[0025] The bandwidth compression ratio is calculated based on the transmission predicted bandwidth. The theoretical maximum bandwidth is used as a benchmark and compared with the predicted transmission bandwidth to calculate the bandwidth compression ratio. The calculation formula is bandwidth compression ratio = 1-(transmission predicted bandwidth / theoretical maximum bandwidth). Through this calculation, the bandwidth compression ratio is obtained. The theoretical maximum bandwidth is pre-set by technical experts.

[0026] Step S200: for the target transmission image, by determining the bandwidth compression ratio, generating an image quality adjustment instruction, wherein a preset compression ratio is used as a determination criterion.

[0027] In the embodiment of the present application, for the target transmission image, the calculated bandwidth compression ratio is compared with the preset compression ratio, wherein the preset compression ratio is a reference value set according to the minimum network requirement for image transmission. If the bandwidth compression ratio is lower than the preset compression ratio, it indicates that the current network bandwidth is insufficient to support normal image quality transmission, and therefore image quality adjustment is required, in which case an image quality adjustment instruction is generated.

[0028] Step S300: Perform adversarial network training with low-order-high-order resolution to obtain a high-order reconstructor, wherein time resolution, spatial resolution and color resolution are used as adjustment targets, and the high-order reconstructor is the generation part of the generative-discriminative architecture retained after training.

[0029] In an embodiment of the present application, adversarial network training is performed with the goal of changing the resolution from low to high order, a generative-discriminative architecture is constructed, and a low-order resolution image is used as input and a high-order target image is used as an output target. During the training process, with time resolution, spatial resolution, and color resolution as adjustment targets, the generative architecture continuously learns the ability to reconstruct high-resolution images from low-resolution images, and determines the difference between the generated image and the target image through the discriminative architecture, iteratively optimizing the generation process. Finally, after the training converges, the generation part in the generative-discriminative architecture is retained as a high-order reconstructor.

[0030] Furthermore, in the method provided in the embodiment of the application, obtaining a high-order reconstructor also includes:

[0031] Construct a generative-discriminative architecture and obtain reconstructed samples, wherein the reconstructed samples include degraded image samples and image samples; determine an initialization generative image of the generative architecture based on the reconstructed samples, discriminate between the initialization generative image and the image samples based on the discriminative architecture, and determine a discriminative loss relationship; adjust the parameters of the generative architecture based on the discriminative loss relationship to obtain a second generative image, and discriminate between the second generative image and the image samples based on the discriminative architecture; perform iterative alternating training until the generative architecture meets the convergence conditions, decouple the generative-discriminative architecture, and use the trained generative architecture as the high-order reconstructor.

[0032] In an embodiment of the present application, a generative adversarial network (GAN) architecture is first constructed, which includes a generative architecture and a discriminative architecture, that is, a generative-discriminative architecture is constructed. The generative architecture is implemented using a convolutional neural network (CNN), which extracts local features of the image through a convolutional layer, and adds a deconvolution layer (upsampling) at the end of the network to improve the resolution of the output image. The discriminative architecture is a binary classification model, whose input can be a generated image or a target high-resolution image, and the output is a judgment value of a real image or a generated image, which is used to calculate the difference between the generated image and the target image.

[0033] In the data preparation stage, the high-resolution image samples are degraded to generate the reconstructed samples required for training. Specifically, Gaussian noise is added by random noise addition to simulate the noise interference of low-quality images; the spatial resolution of the original image is reduced by downsampling; or blurred convolution is used to simulate the defocus or blur effect. The degraded image samples are used as the input of the generation architecture, and the target high-resolution image is used as the output target. This pair of data constitutes the reconstructed samples. The reconstructed samples include degraded image samples-image samples.

[0034] The training process starts with the generative architecture generating the initial image. The degraded image sample is input into the generative architecture, and the generated initial image is used as the generated result; at the same time, the discriminative architecture receives the generated image and the high-resolution target image, and uses the cross-entropy loss function to calculate the discriminative loss relationship between the generated image and the target image. The cross-entropy loss function learns the ability of the discriminator to distinguish between real images and generated images, reflecting the quality gap of the current generated image.

[0035] According to the discriminant loss, the parameters of the generative architecture are adjusted through the back-propagation method to optimize the output effect of the generated image. In this process, the parameters of the generative architecture are updated using the Adam optimizer so that it can gradually generate high-quality outputs that are closer to the target image. The adjusted generative architecture generates a second generated image and inputs it into the discriminant architecture for evaluation again. The discriminant architecture continues to calculate new discriminant losses based on the updated image samples, further providing feedback to the generative architecture.

[0036] Through alternating training of the generative architecture and the discriminative architecture, the generative architecture gradually learns how to restore high-resolution images from degraded image samples. The discriminative architecture improves its ability to distinguish between generated images and real images through continuous optimization, forcing the generative architecture to continuously improve the generation effect. During the training process, the generative architecture gradually masters the ability to restore temporal resolution (frame rate details), spatial resolution (clarity details), and color gradation resolution (color levels and contrast) from low-quality images.

[0037] When the training reaches convergence conditions, that is, the difference between the generated image and the target high-resolution image falls to an acceptable range, and the image quality indicators (such as peak signal-to-noise ratio and structural similarity) are stable, the adversarial training is stopped. At this time, the generative-discriminative architecture is decoupled, and the trained generative architecture is retained as a high-order reconstructor.

[0038] Furthermore, in the method provided in the embodiment of the application, the obtaining of the reconstructed sample further includes:

[0039] Degradation processing is performed on the image samples to determine degraded image samples, wherein random noise is used as the image degradation method; a mapping between the degraded image samples and the image samples is established as the reconstructed samples; wherein the degraded image samples are the initialization generated images of the generation architecture.

[0040] In an embodiment of the present application, high-quality image samples with high resolution and clarity are first extracted from a preset database as target data for training. Next, these high-resolution image samples are degraded to generate degraded image samples similar to actual low-quality scenes. The degradation process uses random noise addition, that is, randomly distributed Gaussian noise is introduced into the pixel values ​​of the original image to simulate the interference that may occur during actual transmission or storage. In this way, the degraded image samples retain the main features of the original image, but the quality is significantly reduced.

[0041] Then, a one-to-one mapping relationship is established between the degraded image samples and the corresponding high-resolution image samples, and the reconstructed samples are determined through this process. This mapping relationship clarifies the relationship between the degraded image samples and the target image samples, that is, the degraded image is used as the input and the high-resolution image is used as the target output to form the reconstructed samples. The degraded image samples are used as the initialization image input of the generative architecture, providing the starting point for the training of the generative architecture and the basis for learning the low-quality to high-quality mapping.

[0042] Step S400: Based on the image quality adjustment instruction, activate the high-order reconstructor, perform high-order reconstruction and loss compensation on the target transmission image, and determine the adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation.

[0043] In an embodiment of the present application, first, a high-order reconstructor is used to preliminarily reconstruct the target transmission image to improve its resolution and overall quality. Subsequently, the brightness and contrast of the reconstructed transmission image are determined based on low light intensity, and a low-light compensation instruction is generated. Then, a high-order curve for low-light compensation is introduced, and the light and dark areas of the image are enhanced according to the low-light compensation instruction, while maintaining the contrast consistency of adjacent pixels to avoid image distortion caused by excessive adjustment. Subsequently, through the introduction of content loss features, content loss compensation is performed on the core areas of the transmission image (such as edges, textures, and patterns) to further optimize the image details and feature restoration. Through the comprehensive processing of high-order reconstruction and low-light compensation and content loss compensation, finally, through the aforementioned process, the adjusted transmission image is determined.

[0044] Furthermore, in the method provided in the embodiment of the application, the loss compensation includes low light compensation and content loss compensation, and further includes:

[0045] The reconstructed transmission image is judged based on low light intensity, and a low light compensation instruction is generated, wherein the judgment standard is a low light intensity threshold; a high-order curve for low light compensation is introduced, and the low light compensation instruction is received to perform light and dark contrast compensation on the reconstructed transmission image, wherein the high-order curve is used to maintain the contrast of adjacent pixels; a content loss feature is introduced to perform content loss compensation on the core feature of the reconstructed transmission image.

[0046] In an embodiment of the present application, the reconstructed transmission image is first judged based on low light levels, that is, by analyzing the brightness values ​​of the pixels of the reconstructed transmission image and comparing them with a preset low light threshold, areas of insufficient brightness are identified, and corresponding low light compensation instructions are generated to identify image areas that require brightness enhancement.

[0047] Next, the high-order curve of low-light compensation is introduced, and the low-light compensation instruction is received to compensate the brightness and darkness contrast of the reconstructed transmission image. First, the brightness difference analysis is performed on the adjacent pixels in the low-light area, and the brightness change value of each pixel and its neighboring pixels is calculated using the sliding window technology, and the contrast enhancement condition, that is, the amplitude modulation range of the brightness adjustment, is set according to the change value. Subsequently, the low-light area in the image is traversed, the actual brightness characteristics are matched with the contrast enhancement conditions one by one, and the pixels that meet the conditions are screened out, and finally the most suitable target enhancement condition for the current image is determined. Based on the target enhancement condition, the brightness analysis is performed on different color partitions of the image, and the boundary contrast features of each partition are extracted, that is, the brightness difference and boundary change between partitions. Combined with these features, the brightness adjustment requirements of each partition are optimized at multiple levels, and the high-order curve of low-light compensation is integrated to generate. This curve is used to control the brightness adjustment amplitude of pixels in the low-light area, while improving the brightness, maintaining the contrast consistency of adjacent pixels, and avoiding the problem of blurred boundaries or unnatural transitions caused by excessive adjustment. During compensation, the high-order curve is used as the adjustment basis, and each pixel in the low-light area is processed one by one. Through this process, the brightness and darkness contrast compensation of the reconstructed transmission image is completed. The high-order curve is used to maintain the contrast of adjacent pixels, that is, the amplitude of pixel adjustment, and the brightness and darkness contrast of the reconstructed transmission image is compensated based on the high-order curve, that is, pixel adjustment.

[0048] After completing low-light compensation, content loss compensation is introduced to further improve the detail performance of the image. First, the core features of the image (such as edges, textures, and color gradients) are analyzed. For edge areas, edge detection techniques such as the Canny algorithm are used to identify blurred or broken boundary details, and these areas are optimized through edge enhancement methods to enhance the clarity of the boundaries and make them sharper. For texture features, the texture distribution in low-light areas is analyzed, and texture generation techniques such as texture filling methods based on surrounding features are used to supplement texture-missing areas to ensure the natural connection between texture and surrounding areas. For areas with abnormal color gradients, the continuity of brightness and color distribution is detected, and color breaks or mutations caused by brightness compensation are identified. Smoothing adjustment techniques such as color interpolation methods are used to repair them to ensure natural and smooth transitions in color and brightness.

[0049] Through the comprehensive processing of low-light compensation and content loss compensation, the reconstructed transmitted image is fully optimized in terms of brightness, contrast and detail performance.

[0050] Furthermore, in the method provided in the embodiment of the application, the high-order curve for introducing low-light compensation further includes:

[0051] According to the image pixel contrast characteristics under various lighting conditions, the contrast enhancement condition is determined, and the contrast enhancement condition is the contrast amplitude modulation of adjacent pixels; the contrast enhancement conditions are traversed, the reconstructed transmission image is matched, and the target enhancement condition is determined; based on the boundary contrast characteristics of the reconstructed transmission image, the target enhancement condition is adjusted in multiple levels, and the high-order curve is generated by integration, wherein the boundary contrast characteristics are the boundary contrast standards of different color scale partitions of the reconstructed transmission image, and the multi-level adjustment is the adjustment requirement of different color scale partitions.

[0052] In the embodiment of the present application, for the contrast characteristics of image pixels under various lighting conditions, the pixel brightness distribution of the reconstructed transmission image is first analyzed, and the brightness values ​​of each pixel and its adjacent pixels are calculated one by one using the sliding window technology to measure the brightness difference between pixels. This brightness difference is used to describe the contrast characteristics between pixels, that is, the regularity of brightness changes in a local area. The sliding window technology defines a local area of ​​a fixed range around each pixel, and calculates the brightness difference changes of all pixels in the area pixel by pixel, ensuring that the analysis results can reflect the local brightness characteristics of the image under different lighting conditions.

[0053] According to the actual needs of various lighting conditions (such as dark light, strong light or uneven light), technical experts set corresponding contrast enhancement conditions, which are used to define the range of brightness changes allowed between pixels, that is, the contrast amplitude modulation of adjacent pixels. For example, under strong light conditions, contrast enhancement conditions tend to limit excessive changes in brightness to avoid overexposure; while under dark light conditions, contrast enhancement conditions allow a larger brightness increase to enhance the contrast effect. Through this analysis process, the preliminary contrast enhancement condition setting is completed.

[0054] Then, all the set contrast enhancement conditions are traversed to match the reconstructed image pixel by pixel. The specific method is to compare the actual brightness difference of each pixel with the enhancement condition to check whether the pixel meets the enhancement rule. In the matching process, the pixels and their local areas that meet the conditions are screened out, and the characteristics of these areas are marked, and finally the target enhancement condition of the current image is determined. The target enhancement condition is an optimization rule generated for the brightness distribution characteristics of the current image to ensure that the enhancement effect adapts to the overall illumination characteristics of the current image.

[0055] Based on the determined target enhancement conditions, the boundary contrast features of the image are further analyzed. The boundary contrast feature refers to the brightness difference at the boundaries of different color-level partitions in the image, which is an important factor affecting the visual layering and natural transition of the image. First, the boundary areas of each color-level partition in the image are extracted through edge detection technology, such as the Canny algorithm. Then, the average brightness difference of the pixels on both sides of the boundary is calculated, and the brightness change characteristics of the boundary are extracted by combining the shape characteristics of the boundary area, such as edge length, edge smoothness, and brightness change directionality. These characteristics are converted into boundary contrast standards to quantify the contrast adjustment requirements of the partition boundaries. For example, for high brightness difference boundaries, emphasis is placed on maintaining contrast clarity; for low brightness difference boundaries, larger brightness adjustments are allowed to enhance the details of the transition area; for transition area boundaries, smooth adjustments are prioritized to ensure natural brightness and color transitions between regions.

[0056] Based on the boundary contrast standard, the target enhancement conditions are adjusted in multiple levels. Specifically, the brightness adjustment strategies are set for different color-level partitions of the image according to the boundary contrast characteristics. For example, the brightness increase range is limited in the high brightness difference area to protect the boundary clarity, the low brightness difference area allows a wider range of contrast enhancement to highlight the details, and the transition area adopts a moderate brightness adjustment range and focuses on smoothing. Through multi-level adjustment, the brightness enhancement effect of different partitions is optimized, while avoiding boundary blur or color distortion caused by overcompensation.

[0057] Finally, the results of multi-level adjustments are integrated to generate a high-order curve that adapts to the entire image.

[0058] Step S500: Visualizing the adjusted transmission image on a sink display interface.

[0059] In an embodiment of the present application, the adjusted transmission image is sent to the sink device through a data transmission link and directly displayed on the sink display interface. During the transmission process, the image data is sent using an established transmission protocol, such as TCP / IP or UDP. After the sink device receives the image, it directly maps the adjusted transmission image to the display interface in pixel order to complete the visualization of the adjusted transmission image.

[0060] Furthermore, the method provided in the application embodiment also includes:

[0061] If it is a continuous frame image, the image blocks are divided by boundary recognition, wherein each image block contains at least one frame rate image, and the consistency of the neighborhood pixels is used as the division standard; for the first image block, the first frame rate image is reconstructed by identifying the edge migration part, wherein the edge migration part is marked with a color scale division boundary, wherein the edge migration part is the migration part of the second frame rate image compared to the first frame rate image, and the color scale division boundary is the color scale dividing line within the edge migration part; based on the first frame rate image, the edge migration part is expanded to determine the second frame rate image.

[0062] In an embodiment of the present application, for the processing of continuous frame images, the image is first divided into multiple image blocks by using boundary recognition technology to reduce the amount of calculation for frame-by-frame processing. Specifically, the brightness or color value of the image is analyzed pixel by pixel using a sliding window technique to calculate the brightness difference or color difference between each pixel and its neighboring pixels. If the difference is less than a preset consistency threshold, the pixels are classified into the same area, thereby dividing a group of image blocks with similar characteristics. Each image block contains at least one frame rate image.

[0063] Among the divided image blocks, the first image block is selected as the processing target, and the first frame rate image therein is reconstructed. First, the Canny operator is used to detect the edge of the image and extract the boundary contour of the object. These edges are usually located in the area where the brightness or color gradient changes most significantly. Then, by further analyzing the pixel motion vectors in the edge area, the dense optical flow method is used to identify the edge migration part, that is, the pixel area in the second frame rate image that is displaced compared to the first frame rate image. Next, the brightness or color characteristics of the pixels in the edge migration part are analyzed, the brightness difference or color difference of the pixels on both sides of the boundary is calculated, and the color level division boundary, that is, the dividing line formed by the brightness or color change, is identified. These boundaries provide standards for color and brightness adjustment for the reconstruction of the first frame rate image.

[0064] After completing the reconstruction of the first frame rate image, it is extended to the second frame rate image based on the determined color scale division boundary. Specifically, the edge migration part of the second frame rate image is extracted by edge detection, and the brightness and color characteristics of the corresponding migration part in the first frame rate image are matched pixel by pixel. For areas with consistent color scale characteristics on both sides of the boundary, the processing results of the areas with the same color scale characteristics in the first frame rate image are directly extended to the second frame rate image to avoid repeated calculations. If it is found that the boundary characteristics have changed significantly, such as the brightness difference exceeds the threshold or the color characteristics are inconsistent, these changed areas are reconstructed and the color scale boundaries are re-divided to adapt to the new frame rate characteristics.

[0065] Finally, the expansion result is integrated with the recalculated change area to generate a complete second frame rate image. During the integration process, the transition between the expanded area and the reconstructed area is ensured to be smooth, and the boundaries are clear and natural, so as to achieve high-quality reconstruction of the second frame rate image. With this method, only the first frame rate image needs to be fully processed, and subsequent frame rate images can be quickly generated based on boundary consistency and color scale expansion technology, which greatly reduces the computational overhead of frame-by-frame reconstruction while ensuring the visual coherence and detail fidelity of the image in dynamic scenes. Finally, this method efficiently handles the dynamically changing areas in continuous frame images, improving the overall processing efficiency and image quality.

[0066] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects:

[0067] The present application uses a time series model to predict network bandwidth and determine a predicted transmission bandwidth, wherein the predicted transmission bandwidth is identified by a bandwidth compression ratio, wherein a target transmission image from a source to a destination is an image transmission task; for the target transmission image, by discriminating the bandwidth compression ratio, a picture quality adjustment instruction is generated, wherein a preset compression ratio is used as a judgment criterion; adversarial network training is performed with a low-order-high-order resolution to obtain a high-order reconstructor, wherein time resolution, spatial resolution and color resolution are used as adjustment targets, and the high-order reconstructor is a generation part of a generation-discrimination architecture retained after training; based on the picture quality adjustment instruction, a high-order reconstructor is activated, and a high-order reconstruction and loss compensation are performed on the target transmission image to determine an adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation; and the adjusted transmission image is visualized on a destination display interface. The present invention solves the technical problems in the prior art of reduced image transmission quality and low transmission efficiency under network bandwidth fluctuations. The network bandwidth is predicted by a time series model, image quality adjustment instructions are generated, and an adversarial network is used to obtain a high-order reconstructor. The transmitted image is reconstructed in a high-order manner and compensated for low light and content loss. Finally, the adjusted image is output as a visualization, thereby achieving the technical effect of realizing high-quality image transmission and improving transmission efficiency under conditions of unstable network bandwidth.

[0068] Embodiment 2 is based on the same inventive concept as the method for adjusting image quality by adaptive network bandwidth in the above embodiment. Figure 2 As shown, the present application provides an image quality adjustment system for adaptive network bandwidth, and the system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0069] A network bandwidth prediction module 11, wherein the network bandwidth prediction module 11 uses a time series model to perform network bandwidth prediction and determine a transmission prediction bandwidth, wherein the transmission prediction bandwidth is marked with a bandwidth compression ratio, wherein a target transmission image from a source to a destination is an image transmission task; an image quality adjustment instruction generation module 12, wherein the image quality adjustment instruction generation module 12 generates an image quality adjustment instruction for a target transmission image by determining the bandwidth compression ratio, wherein a preset compression ratio is used as a judgment standard; a high-order reconstruction acquisition module 13, wherein the high-order reconstruction acquisition module 13 performs reconstruction based on a low-order-high-order resolution. Resistant network training to obtain a high-order reconstructor, wherein time resolution, spatial resolution and color resolution are used as adjustment targets, and the high-order reconstructor is the generation part of the generative-discriminative architecture retained after training; an adjusted transmission image determination module 14, wherein the adjusted transmission image determination module 14 activates the high-order reconstructor based on the image quality adjustment instruction, performs high-order reconstruction and loss compensation on the target transmission image, and determines an adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation; a visualization module 15, wherein the visualization module 15 visualizes the adjusted transmission image on a sink display interface.

[0070] Furthermore, the system is also used to implement the following functions:

[0071] For the target transmission image, a transmission path from a source to a destination is obtained, wherein the transmission path is marked with a transmission time zone; using the time series model, a network bandwidth prediction is performed on the transmission path to determine the transmission prediction bandwidth; based on the transmission prediction bandwidth, the bandwidth compression ratio is calculated.

[0072] Furthermore, the system is also used to implement the following functions:

[0073] Construct a generative-discriminative architecture and obtain reconstructed samples, wherein the reconstructed samples include degraded image samples and image samples; determine an initialization generative image of the generative architecture based on the reconstructed samples, discriminate between the initialization generative image and the image samples based on the discriminative architecture, and determine a discriminative loss relationship; adjust the parameters of the generative architecture based on the discriminative loss relationship to obtain a second generative image, and discriminate between the second generative image and the image samples based on the discriminative architecture; perform iterative alternating training until the generative architecture meets the convergence conditions, decouple the generative-discriminative architecture, and use the trained generative architecture as the high-order reconstructor.

[0074] Furthermore, the system is also used to implement the following functions:

[0075] Degradation processing is performed on the image samples to determine degraded image samples, wherein random noise is used as the image degradation method; a mapping between the degraded image samples and the image samples is established as the reconstructed samples; wherein the degraded image samples are the initialization generated images of the generation architecture.

[0076] Furthermore, the system is also used to implement the following functions:

[0077] The reconstructed transmission image is judged based on low light intensity, and a low light compensation instruction is generated, wherein the judgment standard is a low light intensity threshold; a high-order curve for low light compensation is introduced, and the low light compensation instruction is received to perform light and dark contrast compensation on the reconstructed transmission image, wherein the high-order curve is used to maintain the contrast of adjacent pixels; a content loss function is introduced to perform content loss compensation on the core features of the reconstructed transmission image.

[0078] Furthermore, the system is also used to implement the following functions:

[0079] According to the image pixel contrast characteristics under various lighting conditions, the contrast enhancement condition is determined, and the contrast enhancement condition is the contrast amplitude modulation of adjacent pixels; the contrast enhancement conditions are traversed, the reconstructed transmission image is matched, and the target enhancement condition is determined; based on the boundary contrast characteristics of the reconstructed transmission image, the target enhancement condition is adjusted in multiple levels, and the high-order curve is generated by integration, wherein the boundary contrast characteristics are the boundary contrast standards of different color scale partitions of the reconstructed transmission image, and the multi-level adjustment is the adjustment requirement of different color scale partitions.

[0080] Furthermore, the system is also used to implement the following functions:

[0081] If it is a continuous frame image, the image blocks are divided by boundary recognition, wherein each image block contains at least one frame rate image, and the consistency of the neighborhood pixels is used as the division standard; for the first image block, the first frame rate image is reconstructed by identifying the edge migration part, wherein the edge migration part is marked with a color scale division boundary, wherein the edge migration part is the migration part of the second frame rate image compared to the first frame rate image, and the color scale division boundary is the color scale dividing line within the edge migration part; based on the first frame rate image, the edge migration part is expanded to determine the second frame rate image.

[0082] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0083] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0084] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for adjusting image quality based on adaptive network bandwidth, characterized in that: The method comprises: Using a time series model, network bandwidth prediction is performed to determine a transmission prediction bandwidth, wherein the transmission prediction bandwidth is identified by a bandwidth compression ratio, wherein a target transmission image from a source to a destination is an image transmission task; For the target transmission image, by determining the bandwidth compression ratio, an image quality adjustment instruction is generated, wherein the preset compression ratio is used as a determination criterion; Conducting adversarial network training with low-order-high-order resolution to obtain a high-order reconstructor, wherein the temporal resolution, spatial resolution and color resolution are adjustment targets, and the high-order reconstructor is the generative part of the generative-discriminative architecture retained after training; Based on the image quality adjustment instruction, activating a high-order reconstructor, performing high-order reconstruction and loss compensation on the target transmission image, and determining an adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation; Visualizing the adjusted transmission image on a sink display interface; Among them, obtaining a high-order reconstructor includes: Constructing a generative-discriminative architecture and obtaining a reconstructed sample, wherein the reconstructed sample includes a degraded image sample-an image sample; Based on the reconstructed sample, determine an initialization generated image of a generation architecture, discriminate between the initialization generated image and the image sample based on a discriminant architecture, and determine a discriminant loss relationship; Based on the discriminant loss relationship, adjusting the parameters of the generation architecture to obtain a second generated image, and discriminating the second generated image from the image sample based on the discriminant architecture; Performing iterative alternating training until the generative architecture meets the convergence condition, decoupling the generative-discriminative architecture, and using the trained generative architecture as the high-order reconstructor; Wherein, obtaining the reconstruction sample comprises: Performing degradation processing on the image samples to determine degraded image samples, wherein random noise addition is used as the image degradation method; Establishing a mapping between the degraded image samples and the image samples as the reconstructed samples; Wherein, the degraded image sample is an initialization generated image of the generated architecture; The loss compensation includes low light compensation and content loss compensation, including: The reconstructed transmission image is judged based on low light intensity, and a low light compensation instruction is generated, wherein the judgment standard is a low light intensity threshold; Introducing a high-order curve for low-light compensation, receiving the low-light compensation instruction, and performing light-dark contrast compensation on the reconstructed transmission image, wherein the high-order curve is used to maintain the contrast of adjacent pixels; A content loss function is introduced to compensate for the content loss of the core features of the reconstructed transmitted image.

2. The method for adjusting image quality based on adaptive network bandwidth as claimed in claim 1, characterized in that: The determining of the transmission prediction bandwidth comprises: For the target transmission image, obtaining a transmission path from a source to a destination, wherein the transmission path is marked with a transmission time zone; Using the time series model, predicting the network bandwidth of the transmission path to determine the transmission prediction bandwidth; The bandwidth compression ratio is calculated based on the transmission predicted bandwidth.

3. The method for adjusting image quality based on adaptive network bandwidth as claimed in claim 1, characterized in that: The high-order curve for introducing low-light compensation includes: Determining a contrast enhancement condition for image pixel contrast characteristics under various lighting conditions, wherein the contrast enhancement condition is contrast amplitude modulation of adjacent pixels; Traversing the contrast enhancement conditions, matching the reconstructed transmission image, and determining a target enhancement condition; Based on the boundary contrast characteristics of the reconstructed transmission image, the target enhancement conditions are adjusted in multiple levels and integrated to generate the high-order curve, wherein the boundary contrast characteristics are the boundary contrast standards of different color scale partitions of the reconstructed transmission image, and the multiple levels of adjustment are the adjustment requirements of different color scale partitions.

4. The method for adjusting image quality according to adaptive network bandwidth as claimed in claim 1, characterized in that: The method further comprises: If it is a continuous frame image, the image blocks are divided by boundary recognition, wherein each image block contains at least one frame rate image, and the consistency of the neighboring pixels is used as the division standard; For the first image block, reconstructing the first frame rate image by identifying an edge migration portion, wherein the edge migration portion is marked with a color scale division boundary, wherein the edge migration portion is a migration portion of the second frame rate image compared to the first frame rate image, and the color scale division boundary is a color scale division line within the edge migration portion; Based on the first frame rate image, the edge migration portion is extended to determine a second frame rate image.

5. An image quality adjustment system for adaptive network bandwidth, characterized in that: A system for executing the method for adjusting image quality of an adaptive network bandwidth as claimed in any one of claims 1 to 4, comprising: A network bandwidth prediction module, wherein the network bandwidth prediction module uses a time series model to perform network bandwidth prediction and determine a transmission prediction bandwidth, wherein the transmission prediction bandwidth is identified by a bandwidth compression ratio, wherein a target transmission image from a source to a destination is an image transmission task; A picture quality adjustment instruction generation module, wherein the picture quality adjustment instruction generation module generates a picture quality adjustment instruction for a target transmission image by determining the bandwidth compression ratio, wherein a preset compression ratio is used as a determination criterion; A high-order reconstructor acquisition module, wherein the high-order reconstructor acquisition module performs adversarial network training with low-order-high-order resolution to obtain a high-order reconstructor, wherein time resolution, spatial resolution and color resolution are adjustment targets, and the high-order reconstructor is a generation part of a generative-discriminative architecture that retains training; an adjusted transmission image determination module, wherein the adjusted transmission image determination module activates a high-order reconstructor based on the image quality adjustment instruction, performs high-order reconstruction and loss compensation on the target transmission image, and determines an adjusted transmission image, wherein the loss compensation includes low-light compensation and content loss compensation; A visualization module is used to visualize the adjusted transmission image on a sink display interface.

Images