Content-aware based adaptive regulation method and system for LED display screen

Abstract

This application provides a content-aware adaptive control method and system for LED displays. The method involves multi-scale feature extraction and channel attention recalibration of the video frame to be displayed, resulting in a perceptual feature map with differentiated channel weights. A semantic segmentation network is used to perform semantic segmentation on the perceptual feature map, obtaining multiple semantic category groups for pixel division in the video frame to be displayed, and determining the semantic category label for each semantic category group. Based on all semantic category labels and a preset semantic-visual mapping table, a sensitivity function parameter corresponding to human eye contrast is matched for each semantic category group. Then, based on all sensitivity function parameters and the current ambient light intensity, partition control parameters for different semantic category groups are generated. The partition control parameters are input to the display's driving circuit for content-adaptive partition display control. Based on the above scheme, fine-grained partition control of the display can be achieved based on content semantic awareness and ambient light adaptation.

Description

Technical Field

[0001] This application relates to the field of display device manufacturing technology, and more specifically, to a content-aware adaptive control method and system for LED displays. Background Technology

[0002] As the core display terminal for outdoor media, security monitoring, and commercial displays, the display effect of LED displays is affected by the characteristics of the displayed content, the intensity of ambient light, and the visual perception of the human eye. In order to balance display quality, viewing comfort, and energy consumption control, the industry urgently needs a refined display control solution with content perception and environmental adaptability.

[0003] Existing LED display control technologies mostly adopt a globally unified approach to adjust brightness and gamma correction curves. They do not perform semantic partitioning recognition of the displayed content, nor do they combine multi-scale feature extraction, channel attention recalibration, and ambient light perception to generate differentiated parameters. Furthermore, they do not optimize the control logic to match the contrast sensitivity characteristics of the human eye. As a result, they suffer from poor display effects of key content, weak adaptation to ambient light, poor visual experience, and an inability to coordinate the optimization of energy consumption and image quality. Therefore, how to achieve fine-grained partitioning control of LED displays based on content semantic perception and ambient light adaptation has become a challenge for the industry. Summary of the Invention

[0004] This application provides a content-aware adaptive control method and system for LED displays, which can realize fine-grained control of LED display zones based on content semantic awareness and ambient lighting adaptation.

[0005] In a first aspect, this application provides a content-aware adaptive control method for LED displays, comprising the following steps: The video frame to be displayed is acquired, and multi-scale feature extraction and channel attention recalibration are performed on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights. The perceptual feature map is semantically segmented by a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and the semantic category label of each semantic category group is determined. The light intensity of the current environment is obtained, and the corresponding human eye contrast sensitivity function parameters are matched for each semantic category group according to all semantic category labels and the preset semantic-visual mapping table. Then, based on all the sensitivity function parameters and the light intensity, partition control parameters for different semantic category groups are generated. The zoning control parameters are input into the driving circuit of the LED display screen for content adaptive zoning display control.

[0006] In some embodiments, performing multi-scale feature extraction and channel attention recalibration on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights specifically includes: The video frame to be displayed is input into the encoder of the semantic segmentation network for multi-scale feature extraction to obtain a multi-channel feature map; The multi-channel feature map is compressed by global average pooling using a channel attention mechanism to obtain a compressed vector in the channel dimension. The compressed vector is recalibrated with channel attention to obtain a perceptual feature map with differentiated channel weights.

[0007] In some embodiments, the perceptual feature map is semantically segmented using a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and the semantic category label of each semantic category group is specifically determined as follows: The perceptual feature map is aggregated using a semantic segmentation network with hollow spatial pyramid pooling to obtain a feature tensor that integrates multi-scale semantic information. The feature tensor is upsampled and the low-level features of the corresponding level are fused through skip connections to obtain a semantic probability map with the same resolution as the video frame to be displayed. The pixels in the video frame to be displayed are divided into multiple semantic category groups based on the semantic category probability of each pixel in the semantic probability map. Determine the semantic category label for each semantic category group.

[0008] In some embodiments, matching the corresponding human eye contrast sensitivity function parameters for each semantic category group based on all semantic category labels and a preset semantic-visual mapping table specifically includes: For each semantic category group, the pre-defined semantic-visual mapping table is queried based on the semantic category label of the semantic category group to obtain the visual saliency weight of the semantic category group; Based on the visual saliency weight, a preset human eye contrast sensitivity function model is invoked and the sensitivity curve is adjusted by weighting to obtain the human eye contrast sensitivity function parameters corresponding to the semantic category group, and then the human eye contrast sensitivity function parameters corresponding to each semantic category group are obtained.

[0009] In some embodiments, generating partition control parameters for different semantic category groups based on all sensitivity function parameters and the illumination intensity specifically includes: The brightness compensation coefficient for each semantic category group is determined based on the light intensity. The sensitivity function parameters and brightness compensation coefficients of each semantic category group are weighted and fused to generate the brightness gain coefficient and gamma correction curve corresponding to each semantic category group. Based on the luminance gain coefficients and gamma correction curves of all semantic category groups, determine the partition control parameters for different semantic category groups.

[0010] In some embodiments, inputting the zoning control parameters to the driving circuit of the LED display screen for content-adaptive zoning display control specifically includes: The zoning control parameters are transmitted to the drive circuit through the bus interface of the LED display control terminal; The driving circuit performs pixel-level brightness scaling on the video frame to be displayed according to the brightness gain coefficient of each semantic category group in the partition control parameters. The driving circuit performs a non-linear mapping on the scaled pixel values ​​in the video frame to be displayed according to the gamma correction curve of each semantic category group in the partition control parameters, and obtains the corrected driving level. The driving circuit generates a pulse width modulation signal with a corresponding duty cycle according to the driving level and applies it to the light-emitting diodes in the LED display screen corresponding to each semantic category group, so that each semantic category group presents the display content with differentiated brightness and contrast.

[0011] In some embodiments, the semantic category label includes a portrait area, a text area, and a background area.

[0012] Secondly, this application provides a content-aware adaptive control system for LED displays, used to execute a content-aware adaptive control method for LED displays, including: The acquisition module is used to acquire video frames to be displayed, and to perform multi-scale feature extraction and channel attention recalibration on the video frames to be displayed to obtain a perceptual feature map with differentiated channel weights. The processing module is used to perform semantic segmentation on the perceptual feature map through a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and to determine the semantic category label of each semantic category group. The processing module is also used to obtain the light intensity of the current environment, and match the corresponding human eye contrast sensitivity function parameters for each semantic category group according to all semantic category labels and the preset semantic-visual mapping table, and then generate partition control parameters for different semantic category groups according to all sensitivity function parameters and the light intensity. The execution module is used to input the partition control parameters to the driving circuit of the LED display screen for content adaptive partition display control.

[0013] Thirdly, this application provides a computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described content-aware adaptive control method for LED displays.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described content-aware adaptive control method for LED displays.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application provides a content-aware adaptive control method and system for LED displays, which involves acquiring a video frame to be displayed and performing multi-scale feature extraction and channel attention recalibration on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights. The perceptual feature map is semantically segmented using a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and the semantic category label of each semantic category group is determined. The illumination intensity of the current environment is obtained, and the corresponding human eye contrast sensitivity function parameters are matched for each semantic category group according to all semantic category labels and a preset semantic-visual mapping table. Then, partition control parameters for different semantic category groups are generated according to all sensitivity function parameters and the illumination intensity. The partition control parameters are input to the driving circuit of the LED display screen for content adaptive partition display control.

[0016] Therefore, in this application, the partition control parameters are input to the driving circuit of the LED display screen for content-adaptive partition display control. First, by determining the semantic category group, connected regions formed by pixel clusters with the same semantic category and adjacent spatial positions can be obtained. In this scheme, the semantic category group determined by multi-scale feature extraction, channel attention recalibration, and semantic segmentation can accurately divide the video frame to be displayed into regions with clear semantic attributes such as portraits, text, and backgrounds by pixels. This provides a precise partitioning basis for subsequent matching of the semantic-visual mapping table and obtaining the corresponding human eye contrast sensitivity function parameters. This not only breaks the limitation of the existing technology of global unified control, realizing the transformation of display control from global coarse adjustment to semantic partition fine adjustment, but also generates differentiated partition control parameters based on ambient light intensity. Then, by determining the semantic category label, the semantic category identifier assigned to each semantic category group can be obtained. The semantic category label can accurately identify the core visual attributes of each semantic category group, providing a basis for querying the semantic-visual mapping table and matching the corresponding human eye contrast sensitivity function. The parameters provide a unique and accurate index, directly binding the generation of control parameters to the semantic importance of the content. This completely breaks through the limitations of traditional global coarse adjustment, allowing the system to generate differentiated zone control parameters based on ambient light intensity. Prioritizing the display clarity and contrast of high visually important semantic areas, while conforming to the laws of human visual perception, it significantly improves overall display quality and viewing comfort, achieving synergistic optimization of display quality and energy consumption. This effectively solves the core problems of poor display of key content, weak adaptability to ambient light, and poor visual experience in existing technologies. Finally, determining the zone control parameters provides control instructions for the LED display driver circuit to perform differentiated brightness scaling and grayscale correction on each display zone. The zone control parameters are generated based on the semantic category labels of each semantic category group, matching the human eye's contrast sensitivity function parameters, and integrating the current ambient light intensity. This allows for the output of differentiated brightness gain coefficients and gamma correction curves for different semantic areas such as portraits, text, and backgrounds, completely breaking through the limitations of traditional global coarse adjustment. The display driving circuit can achieve precise zone control based on the semantic importance of the content and ambient lighting conditions, prioritizing the display clarity and contrast of visually important areas. At the same time, it conforms to the visual perception rules of the human eye, significantly improving display quality and viewing comfort while achieving synergistic optimization of image quality and energy consumption, effectively solving the core problems of existing technologies. In summary, based on the above solution, it is possible to achieve fine-grained zone control of LED displays based on content semantic perception and ambient lighting adaptation. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of a content-aware adaptive control method for LED displays, as shown in some embodiments of this application. Figure 2This is an exemplary flowchart illustrating the determination of a perceived feature map according to some embodiments of this application; Figure 3 This is a schematic diagram illustrating the application scenario of the semantic segmentation network in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of a content-aware adaptive control system for an LED display screen, as shown in some embodiments of this application. Figure 5 This is a schematic diagram of the structure of a computer device that implements a content-aware adaptive control method for LED displays, according to some embodiments of this application. Detailed Implementation

[0018] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figure 1 The figure is an exemplary flowchart of a content-aware adaptive control method for LED displays according to some embodiments of this application. The figure mainly includes the following steps: In step 101, a video frame to be displayed is acquired, and multi-scale feature extraction and channel attention recalibration are performed on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights.

[0020] It should be noted that in this application, the video frame to be displayed is image data obtained from the video source without feature extraction processing.

[0021] In some embodiments, acquiring the video frame to be displayed can be achieved in the following manner: First, the original video frame can be acquired in real time through the video stream acquisition interface deployed on the LED display control terminal; then, the acquired original video frame is subjected to resolution adaptation processing, and its size is uniformly adjusted to a preset standard resolution, which matches the size required by the input layer of the convolutional neural network; then, the resolution-adapted original video frame is converted from the first color space to the second color space to separate the luminance component and chrominance component of the image, so that the subsequent multi-scale feature extraction module can focus on the luminance component; finally, the original video frame after resolution adaptation and color space conversion is used as the video frame to be displayed; wherein, the video stream acquisition interface establishes a data connection with the external video source device and can receive the original video stream data at a rate of not less than 24 frames per second; the first color space is the red-green-blue color space; the second color space is the luminance-chrominance color space.

[0022] In some embodiments, reference Figure 2The figure is an exemplary flowchart of determining a perceptual feature map according to some embodiments of this application. In this application, multi-scale feature extraction and channel attention recalibration are performed on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights, which can be achieved by the following steps: In step 1011, the video frame to be displayed is input into the encoder of the semantic segmentation network for multi-scale feature extraction to obtain a multi-channel feature map; In step 1012, the multi-channel feature map is compressed using a channel attention mechanism through global average pooling to obtain a compressed vector in the channel dimension. In step 1013, the compressed vector is recalibrated for channel attention to obtain a perceptual feature map with differentiated channel weights.

[0023] It should be noted that in this application, references Figure 3 This figure illustrates an application scenario of the semantic segmentation network in this application embodiment. The semantic segmentation network is a deep learning model for feature extraction from video frames to be displayed. It consists of four parts: an encoder, a dilated spatial pyramid pooling module, a decoder, and a pixel-wise classification layer. The encoder adopts a dilated convolutional network structure, which achieves multi-scale feature extraction from the video frames to be displayed by setting at least two dilated convolutional kernels with different dilation rates. A compression-excitation network module is embedded in the encoder as a channel attention mechanism to perform global average pooling compression and channel weight recalibration on the extracted multi-channel feature maps. The dilated spatial pyramid pooling module captures the multi-scale context of the feature maps from local details to global semantics by setting multiple dilated convolutional layers with different dilation rates in parallel. Information is fused into a feature tensor. The decoder progressively upsamples the feature tensor and fuses it with low-level features from the corresponding encoder layer through skip connections to recover spatial detail information. The pixel-by-pixel classification layer outputs a semantic probability map with the same resolution as the video frame to be displayed based on the fused features, and assigns a semantic category probability to each pixel. This semantic segmentation network can extract a perceptual feature map with multi-scale contextual information and differentiated channel weights from the video frame to be displayed, and accurately divide the display content into multiple semantic category groups with clear semantic meanings through pixel-level semantic recognition. This provides a fine-grained and high-confidence semantic decision basis for the subsequent generation of partitioning control parameters based on the semantic-visual mapping table, thereby achieving deep coupling between display optimization and content semantics.

[0024] In specific implementation, the video frame to be displayed is input into the encoder of the semantic segmentation network for multi-scale feature extraction to obtain a multi-channel feature map. This can be achieved in the following way: First, the video frame to be displayed can be input into the encoder of the convolutional neural network for multi-scale feature extraction. This encoder adopts a dilated convolutional network structure, and performs convolution operations on the video frame to be displayed by setting three dilated convolutional kernels with different dilation rates. The first dilation rate is set to a value of 1 to capture local detail features, the second dilation rate is set to a value of 2 to capture mid-range contextual information, and the third dilation rate is set to a value of 4 to capture global semantic information. The different dilation rates are then combined. Multiple feature maps extracted by the tensor dilated convolution kernel are concatenated along the channel dimension to obtain an intermediate feature map that integrates multi-scale contextual information. Then, the intermediate feature map is passed sequentially through a batch normalization layer and a modified linear unit activation function layer for feature normalization and nonlinear mapping to obtain a multi-channel feature map with enhanced expressive power. The multi-channel feature map is a three-dimensional data tensor containing multiple independent feature channels. Each channel corresponds to an activation response of a semantic or texture feature, which can provide a feature foundation containing rich spatial and semantic information for subsequent channel attention recalibration, enabling the system to perform differentiated processing based on the importance of different channels.

[0025] In specific implementation, the multi-channel feature map is compressed using a channel attention mechanism through global average pooling to obtain a compressed vector in the channel dimension. This can be achieved as follows: the multi-channel feature map is compressed using a channel attention mechanism of a convolutional neural network through global average pooling to obtain a compressed vector in the channel dimension. This channel attention mechanism adopts a compression-activation network structure. The compression module first performs a global average pooling operation on the multi-channel feature map in the spatial dimension, compressing the spatial feature of each channel into a scalar value. This scalar value represents the global response intensity of the channel feature map across the entire spatial range. For a multi-channel feature map containing 256 channels... The feature map, after global average pooling, yields a compressed vector with a dimension of 256. Each dimension corresponds to the global statistical information of one channel. The compression process reduces the spatial dimension of the feature map (height x width) to a one-dimensional vector, enabling subsequent activation operations to learn the dependencies between channels based on the global receptive field. The compressed vector refers to a one-dimensional statistic with a dimension equal to the number of channels. Each component represents the global response intensity of the corresponding channel across the entire image space. It can aggregate the scattered information of the spatial dimension into a global description of the channel dimension, enabling the channel attention mechanism to learn the dependencies between channels based on the global receptive field rather than local regions.

[0026] In specific implementation, channel attention recalibration of the compressed vector to obtain a perceptual feature map with differentiated channel weights can be achieved in the following way: Channel attention recalibration of the compressed vector can be performed using the channel attention mechanism of a convolutional neural network. This channel attention mechanism adopts a compression-activation network structure, and the activation module consists of two fully connected layers. The first fully connected layer reduces the dimensionality of the compressed vector to one-sixteenth of the original number of channels by introducing nonlinearity through a modified linear unit activation function. The second fully connected layer restores the dimensionality-reduced vector to the original number of channels and normalizes the output value to the interval between zero and one through a gating mechanism function, thus obtaining the activation weights of each channel. Each value in the incentive weights represents the importance of the corresponding channel in subsequent processing. The closer the value is to one, the more important the channel feature is; the closer the value is to zero, the more important the channel feature should be suppressed. The incentive weights are then multiplied one-to-one with the multi-channel feature map initially input to the channel attention mechanism to achieve channel-level feature recalibration. Finally, the recalibrated multi-channel feature map is used as the perceptual feature map. The perceptual feature map refers to a feature map with differentiated weights in the channel dimension, which can enhance the response strength of key feature channels that are semantically related to the displayed content, while suppressing the interference of unimportant feature channels, thus providing more discriminative input features for the subsequent semantic segmentation network.

[0027] In step 102, the perceptual feature map is semantically segmented by a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and the semantic category label of each semantic category group is determined.

[0028] In some embodiments, semantic segmentation of the perceptual feature map is performed using a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and the semantic category label of each semantic category group is determined by the following steps: The perceptual feature map is aggregated using a semantic segmentation network with hollow spatial pyramid pooling to obtain a feature tensor that integrates multi-scale semantic information. The feature tensor is upsampled and the low-level features of the corresponding level are fused through skip connections to obtain a semantic probability map with the same resolution as the video frame to be displayed. The pixels in the video frame to be displayed are divided into multiple semantic category groups based on the semantic category probability of each pixel in the semantic probability map. Determine the semantic category label for each semantic category group.

[0029] In specific implementation, the perceptual feature map is aggregated using dilated spatial pyramid pooling (DSP) through a semantic segmentation network to obtain a feature tensor that fuses multi-scale semantic information. This can be achieved as follows: the perceptual feature map is passed to the DSP module of the semantic segmentation network for feature aggregation. This DSP module contains multiple dilated convolutional layers with different dilation rates, set to 6, 12, 18, and 24 respectively. Each dilated convolutional layer independently extracts features from the perceptual feature map, capturing contextual information at different scales in the image. Specifically, a dilated convolutional layer with a dilation rate of 6 captures small-scale local detail features, while a dilation rate of 24 captures small-scale local detail features. The twelve- and eighteen-level dilated convolutional layers capture mid-scale regional structural features, while the twenty-four-level dilated convolutional layer captures large-scale global semantic features. The feature maps of different scales output from each dilated convolutional layer are then concatenated along the channel dimension. A one-by-one convolutional layer is then used to perform channel dimensionality reduction and feature fusion on the concatenated feature maps, resulting in a feature tensor that integrates multi-scale semantic information. This feature tensor is a multi-dimensional feature representation that simultaneously contains multi-level image information from local details to global semantics of the video frame to be displayed. It provides a unified feature foundation with both spatial resolution and semantic abstraction capabilities for subsequent semantic segmentation, thereby improving the accuracy and robustness of semantic segmentation.

[0030] In specific implementation, upsampling the feature tensor and fusing it with low-level features from the corresponding layer through skip connections to obtain a semantic probability map with the same resolution as the video frame to be displayed can be achieved in the following way: the feature tensor can be passed to the decoder part of the semantic segmentation network. The decoder first performs a progressive upsampling operation on the feature tensor, gradually restoring the spatial resolution of the feature tensor to the same size as the video frame to be displayed. The upsampling factor is determined according to the total downsampling factor of the encoder, usually implemented by bilinear interpolation or transposed convolution. At each stage of upsampling, the decoder obtains low-level feature maps from the corresponding layer of the encoder through skip connections. These low-level feature maps retain rich spatial details such as edges, corners, and textures. The low-level feature maps are then combined with the upsampled feature maps. The images are stitched together along the channel dimension, and then feature fusion is performed through convolutional layers to compensate for the detail information lost during upsampling. After multi-level upsampling and skip connection fusion, the final output is a semantic probability map with the same resolution as the video frame to be displayed. The semantic probability map is a probability distribution map with the same resolution as the video frame to be displayed. Each pixel position in the semantic probability map corresponds to a probability vector with a dimension equal to the total number of preset semantic categories, i.e., semantic category probability. The value of each dimension in the semantic category probability represents the probability value of the pixel belonging to the corresponding semantic category. The preset semantic categories include at least human image category, text category, and background category. It can transform the abstract features extracted by the deep network into interpretable pixel-level semantic prediction results, providing a direct quantitative basis for region division.

[0031] In specific implementation, dividing the pixels in the video frame to be displayed into multiple semantic category groups based on the semantic category probability of each pixel in the semantic probability map can be achieved in the following way: the semantic category probability of each pixel position in the semantic probability map can be analyzed, that is, the index corresponding to the maximum value of the semantic category probability can be extracted, and the semantic category represented by the index can be determined as the semantic category of the pixel. Then, all pixel positions in the semantic probability map are traversed to determine a unique semantic category for each pixel in the video frame to be displayed. Then, pixels with the same semantic category and spatially adjacent positions are clustered into the same connected region. Morphological post-processing is performed on the connected region, including small region culling and hole filling, to eliminate noise interference and segmentation defects. Each post-processed connected region is taken as a semantic category group, thereby dividing all pixels in the video frame to be displayed into multiple non-overlapping semantic category groups. The semantic category group is a connected region formed by clustering pixels with the same semantic category and spatially adjacent positions. It can aggregate discrete pixel-level prediction results into image object regions with complete geometric boundaries and semantic consistency, thereby providing a control granularity that matches the visual perception unit for subsequent partition control.

[0032] It should be noted that in this application, the semantic category label is a semantic category identifier assigned to each semantic category group, including the human image area, text area, and background area. It can clearly distinguish the visual attributes of different semantic category groups, thereby providing an accurate index basis for subsequent querying of the semantic-visual mapping table and matching the corresponding visual saliency weights, and realizing differentiated partitioning control based on content importance. In specific implementation, the semantic category label of each semantic category group can be determined in the following way: For each semantic category group, the probability vectors corresponding to all pixels in the semantic probability map within the semantic category group are counted, and the average probability value of all pixel probability vectors with respect to each semantic category is calculated. Then, the semantic category with the highest average probability value is determined as the semantic category label of the semantic category group. When there is inconsistency in the semantic category identifiers of pixels within the semantic category group, the semantic category of the majority of pixels in the semantic category group is taken as the final semantic category label of the group.

[0033] In step 103, the light intensity of the current environment is obtained, and the corresponding human eye contrast sensitivity function parameter is matched for each semantic category group according to all semantic category labels and the preset semantic-visual mapping table. Then, partition control parameters for different semantic category groups are generated according to all the sensitivity function parameters and the light intensity.

[0034] In some embodiments, the current ambient light intensity can be obtained in the following manner: First, multiple photosensitive sensor arrays deployed around the bezel of the LED display screen can synchronously collect the original analog signal of the current ambient light intensity at a preset sampling frequency. These photosensitive sensor arrays include at least three photosensitive sensors respectively deployed on the top, left, and right bezels of the display screen to eliminate single-point measurement errors caused by local shadows or direct strong light sources. Then, the collected original analog light intensity signal is converted into a digital light intensity sample value using an analog-to-digital converter circuit, and the light intensity from multiple photosensitive sensors within the same sampling period is analyzed. The average illuminance of the current environment is obtained by arithmetic mean filtering of the sampled illuminance values. Finally, the average illuminance is matched with a preset illuminance level lookup table. The illuminance level index corresponding to the current environment is determined based on the range in which the average illuminance falls, and this illuminance level index is used as the illuminance of the current environment. The illuminance is a physical parameter characterizing the brightness of the environment in which the LED display screen is located. It can be used as an input variable for adaptive environmental control to correct display effect deviations caused by ambient light interference. The illuminance level lookup table stores the correspondence between multiple consecutive illuminance value ranges and illuminance level indices.

[0035] In some embodiments, matching the corresponding human eye contrast sensitivity function parameter for each semantic category group based on all semantic category labels and a preset semantic-visual mapping table can be achieved using the following steps: For each semantic category group, the pre-defined semantic-visual mapping table is queried based on the semantic category label of the semantic category group to obtain the visual saliency weight of the semantic category group; Based on the visual saliency weight, a preset human eye contrast sensitivity function model is invoked and the sensitivity curve is adjusted by weighting to obtain the human eye contrast sensitivity function parameters corresponding to the semantic category group, and then the human eye contrast sensitivity function parameters corresponding to each semantic category group are obtained.

[0036] It should be noted that in this application, the semantic-visual mapping table is a pre-defined, rule-based mapping structure used to establish a one-to-one correspondence between semantic category labels and visual saliency weights. This mapping table stores the correspondence between semantic category labels and visual saliency weights in key-value pairs. The semantic category labels include human image regions, text regions, and background regions. The visual saliency weights are normalized values ​​between 0 and 1. The visual saliency weight corresponding to the human image region can be set to 0.9, the visual saliency weight corresponding to the text region can be set to 0.7, and the visual saliency weight corresponding to the background region can be set to 0.3, so as to reflect the difference in the importance of different semantic categories in visual perception. This can transform abstract semantic information into quantifiable visual importance indicators, providing an objective and unified basis for weight allocation for subsequent differentiated control.

[0037] In specific implementation, the visual saliency weight of a semantic category group can be obtained by querying a preset semantic-visual mapping table based on the semantic category label of the semantic category group. This can be achieved in the following way: the preset semantic-visual mapping table can be obtained through the storage unit deployed on the LED display control terminal, and the semantic category label of the semantic category group can be extracted. Then, the semantic category label is used as the query key and input into the semantic-visual mapping table for precise matching. The visual saliency weight uniquely corresponding to the semantic category label is obtained as the visual saliency weight of the semantic category group. The visual saliency weight is a normalized quantitative value that characterizes the relative importance of the semantic category label content in human visual perception. It can realize differentiated resource allocation based on content importance, so that key display content can obtain higher display priority and optimization.

[0038] It should be noted that in this application, the human eye contrast sensitivity function model is a mathematical model that quantitatively describes the human eye's ability to perceive brightness and contrast at different spatial frequencies, with spatial frequency as the independent variable and contrast sensitivity as the dependent variable. In this scheme, it consists of a standard sensitivity curve, a visual saliency weighting adjustment module, and a key feature parameter extraction unit. The standard sensitivity curve uses the Barten model to define the contrast sensitivity characteristics of the human eye under typical conditions. The visual saliency weighting adjustment module scales the amplitude of the standard sensitivity curve according to the visual saliency weights corresponding to the semantic category groups. The key feature parameter extraction unit extracts the peak sensitivity, the spatial frequency corresponding to the peak sensitivity, and the full width at half maximum (FWHM) bandwidth from the adjusted personalized sensitivity curve as sensitivity function parameters. It can transform the content importance index obtained from semantic segmentation into quantitative driving parameters that conform to the physiological characteristics of the human eye, so that the partition control can not only prioritize the display quality of highly important content, but also ensure that the controlled image achieves the optimal contrast presentation effect in human visual perception, thereby realizing the deep integration of content semantic understanding and human visual characteristics.

[0039] In specific implementation, the sensitivity function parameters of the semantic category group corresponding to the human eye contrast are obtained by calling a preset human eye contrast sensitivity function model based on the visual saliency weight and adjusting the sensitivity curve accordingly. This can be achieved in the following way: First, the preset human eye contrast sensitivity function model can be obtained through the storage unit deployed on the LED display control terminal, and the visual saliency weight is input into the human eye contrast sensitivity function model as a weighting coefficient. The amplitude of the standard sensitivity curve is then adjusted by weighting the sensitivity value at each spatial frequency point on the standard sensitivity curve by the visual saliency weight. The corresponding personalized sensitivity curve is obtained; then, key feature parameters are extracted from the personalized sensitivity curve, including peak sensitivity, spatial frequency corresponding to peak sensitivity, and full width at half maximum (FWHM) bandwidth. These key feature parameters are then used to construct the sensitivity function parameters of the human eye contrast corresponding to the semantic category group. The sensitivity function parameters are a set of key feature parameters extracted after weighting the sensitivity curve of the human eye contrast sensitivity function model according to the visual salience weight. This set of parameters can transform the importance of content into a display driving basis that conforms to the physiological characteristics of the human eye, so that the final zoning control further conforms to the visual perception law of the human eye on the basis of content priority and environmental adaptation.

[0040] In some embodiments, generating partition control parameters for different semantic category groups based on all sensitivity function parameters and the illumination intensity can be achieved using the following steps: The brightness compensation coefficient for each semantic category group is determined based on the light intensity. The sensitivity function parameters and brightness compensation coefficients of each semantic category group are weighted and fused to generate the brightness gain coefficient and gamma correction curve corresponding to each semantic category group. Based on the luminance gain coefficients and gamma correction curves of all semantic category groups, determine the partition control parameters for different semantic category groups.

[0041] In specific implementation, determining the brightness compensation coefficient for each semantic category group based on the illumination intensity can be achieved in the following way: First, the sensitivity coefficient of each semantic category group to illumination changes can be determined based on the semantic category label corresponding to each semantic category group. Then, using a preset environmental compensation function, which is a bilinear interpolation function with illumination intensity and sensitivity coefficient as input, the brightness compensation coefficient for each semantic category group is calculated based on the illumination intensity and the sensitivity coefficient of each semantic category group to illumination changes. Among these, the human figure area is the most sensitive to illumination changes, followed by the text area, and the background area is the least sensitive. Therefore, the human figure area corresponds to the first sensitivity coefficient, and the text area corresponds to the second sensitivity coefficient. The coefficient and background area correspond to the third sensitivity coefficient, and the first sensitivity coefficient is greater than the second sensitivity coefficient, which is greater than the third sensitivity coefficient. Furthermore, there is a negative correlation between light intensity and brightness compensation coefficient; that is, the higher the light intensity, the larger the brightness compensation coefficient to offset the contrast loss caused by strong light, and the lower the light intensity, the smaller the brightness compensation coefficient to avoid glare in low-light environments. The brightness compensation coefficient is used to indicate the required brightness enhancement ratio for each semantic category group under the current ambient light intensity, in order to offset the interference of ambient light on display contrast. It enables display control to dynamically adapt to different environmental scenarios from strong light to low light, avoiding the problem of key content being invisible or glare caused by changes in ambient light.

[0042] In specific implementation, the sensitivity function parameters and brightness compensation coefficients of each semantic category group are weighted and fused to generate the corresponding brightness gain coefficient and gamma correction curve for each semantic category group. This can be achieved in the following way: For each semantic category group, the visual saliency weight of the semantic category group can be used as the basic control weight and weighted and fused with the brightness compensation coefficient. That is, the brightness gain coefficient of the semantic category group is obtained by calculating the average value through a preset fusion formula, such as a preset fusion formula. At the same time, the gamma correction curve of the semantic category group is calculated based on the basic control weight and the brightness compensation coefficient through a preset gamma correction mapping function. Here, the basic control weight is used to characterize the visual importance of the corresponding semantic category group at the content level, and the brightness compensation coefficient is used to characterize the brightness requirement of the corresponding semantic category group at the environmental level. The brightness gain coefficient is... The indicator scales the pixel brightness of semantic category groups, enabling a dual-dimensional synergistic drive of content importance and environmental interference. This allows high-importance areas to achieve higher brightness gain in strong light to ensure visibility, and lower brightness gain in low light to improve viewing comfort. The gamma correction curve is a mapping relationship that performs a non-linear transformation of the pixel grayscale values ​​of semantic category groups. The gamma value of this gamma correction curve decreases as the basic control weight increases to enhance the dark detail representation of high-importance areas, and decreases as the brightness compensation coefficient increases to improve overall contrast in strong light environments. It can dynamically adjust grayscale response characteristics according to content importance, allowing high-importance areas to use lower gamma values ​​to enhance dark detail representation, while simultaneously optimizing overall contrast in conjunction with ambient light, significantly improving the quality of human visual perception.

[0043] It should be noted that, in this application, the partition control parameters are control instructions that instruct the LED display driving circuit to perform differentiated brightness scaling and grayscale correction on each display partition; they can convert the semantic understanding results of the content into executable driving parameters, enabling the display to achieve content-adaptive differentiated display control with partition-level precision, improving the visual performance of key content while reducing power consumption in non-critical areas; in specific implementation, the partition control parameters for different semantic category groups can be determined based on the brightness gain coefficients and gamma correction curves of all semantic category groups in the following way: the pixel position coordinates and area range of each semantic category group in the video frame to be displayed can be obtained, and the group brightness gain coefficients and gamma correction curves of each semantic category can be spatially aligned according to all pixel position coordinates and area ranges to form a control parameter mapping table indexed by the semantic category group. Each entry in the control parameter mapping table contains the semantic category label of the semantic category group, pixel position coordinates and area range, corresponding brightness gain coefficient and gamma correction curve; and the control parameter mapping table is then used as the partition control parameters for different semantic category groups.

[0044] In step 104, the partition control parameters are input to the driving circuit of the LED display screen for content-adaptive partition display control.

[0045] In some embodiments, the process of inputting the zoning control parameters into the driving circuit of the LED display screen for content-adaptive zoning display control can be achieved by the following steps: The zoning control parameters are transmitted to the drive circuit through the bus interface of the LED display control terminal; The driving circuit performs pixel-level brightness scaling on the video frame to be displayed according to the brightness gain coefficient of each semantic category group in the partition control parameters. The driving circuit performs a non-linear mapping on the scaled pixel values ​​in the video frame to be displayed according to the gamma correction curve of each semantic category group in the partition control parameters, and obtains the corrected driving level. The driving circuit generates a pulse width modulation signal with a corresponding duty cycle according to the driving level and applies it to the light-emitting diodes in the LED display screen corresponding to each semantic category group, so that each semantic category group presents the display content with differentiated brightness and contrast.

[0046] It should be noted that in this application, the driving circuit is an execution terminal that receives and parses the partition control parameters and converts them into differentiated pulse width modulation signals for different semantic category groups. It can realize precise conduction control of light-emitting diodes, thereby ensuring that the content semantic perception results can be accurately mapped to the physical display level, so that partition control can be transformed from algorithmic decision-making into actual display effect.

[0047] In specific implementation, the transmission of the partition control parameters to the driving circuit through the bus interface of the LED display control terminal can be achieved in the following way: First, the brightness gain coefficient and gamma correction curve of each semantic category group in the partition control parameters can be encapsulated into a control parameter set in a way that corresponds one-to-one with the physical partition of the LED display. The data structure of the control parameter set includes a partition index field, a brightness gain coefficient field, and a gamma correction curve field. Then, the control parameter set is packaged into a data packet conforming to the low voltage differential signal interface protocol by the field programmable gate array chip built into the LED display control terminal, and the data packet is sent to the serial deserializer on the driving circuit side through the low voltage differential signal interface. The serial deserializer restores the received serial data into parallel data and stores it in the random access memory of the driving circuit. Finally, the driving circuit allocates the brightness gain coefficient and gamma correction curve of each partition to the pulse width modulation driving unit corresponding to each partition as the driving parameter reference according to the partition index field stored in the random access memory.

[0048] In specific implementation, the driving circuit performs pixel-level brightness scaling of the video frame to be displayed according to the brightness gain coefficient of each semantic category group in the partition control parameters. This can be achieved in the following way: First, the driving circuit receives the video frame to be displayed and caches it in the frame memory in a data storage structure organized by partition. Second, the timing controller in the driving circuit reads the pixel data of the current partition from the frame memory according to the partition scanning order, and at the same time reads the brightness gain coefficient corresponding to the partition from the random access memory. Then, the pixel processing unit in the driving circuit uses the brightness gain coefficient as a multiplication factor to perform independent multiplication operations on the red channel value, green channel value, and blue channel value of each pixel in the current partition. For example, when the brightness gain coefficient is 1.2, the red channel value is multiplied by 1.2 from the original pixel value of 120 to obtain 144, thus achieving linear scaling of brightness. Finally, the pixel data after brightness scaling is temporarily stored in the row buffer for use in the subsequent gamma correction step.

[0049] In specific implementation, the driving circuit performs non-linear mapping on the scaled pixel values ​​in the video frame to be displayed according to the gamma correction curves of each semantic category group in the partition control parameters to obtain the corrected driving level. This can be achieved in the following way: First, the driving circuit reads the pixel data after brightness scaling from the row buffer and simultaneously reads the gamma correction curve corresponding to the current partition from the random access memory. The gamma correction curve is stored in the form of a lookup table containing 256 mapping nodes, with each node corresponding to a mapping relationship between an input pixel value and an output pixel value. Second, the gamma correction unit in the driving circuit uses the current pixel value as the index address of the lookup table and directly obtains the corrected pixel value through the lookup operation. For example, when the input pixel value is 128, according to the preset gamma value of the correction curve of 2.2, the output pixel value obtained by looking up the table is approximately 106, realizing the non-linear transformation of the pixel value. Then, the driving circuit latches the corrected pixel value as the driving level. The driving level is a digital control quantity after brightness scaling and gamma correction, represented in the form of digital voltage, with a range between 0 and 4095 corresponding to 12-bit grayscale resolution.

[0050] In specific implementation, the driving circuit generates a pulse width modulation signal with a corresponding duty cycle based on the driving level and applies it to the light-emitting diodes in the LED display screen corresponding to each semantic category group, so that each semantic category group presents the display content with differentiated brightness and contrast. This can be achieved in the following way: First, the pulse width modulation generation unit in the driving circuit receives the driving level and loads the value of the driving level into the comparison register of the corresponding pulse width modulation counter. Then, a preset counting period value is loaded into the period register. The counting period value is determined according to the refresh rate of the display screen. For example, when the refresh rate is 60 Hz, the counting period value is set to the number of clock cycles corresponding to 16666 microseconds. Then, the pulse width modulation counter accumulates counts in each clock cycle. When the count value... The output is high when the count value is less than the drive level in the comparison register, and low when the count value is greater than or equal to the drive level, thereby generating a high-frequency pulse width modulation signal with a duty cycle proportional to the drive level. Finally, the pulse width modulation signal is applied to the anode or cathode of the light-emitting diode corresponding to the video frame partition of each semantic category group through the power amplifier circuit of the driver chip, so that the light-emitting diode emits light with a constant current during the conduction time. By adjusting the proportion of conduction time in each pulse width modulation cycle, the differentiated brightness and contrast of each semantic category group are achieved. The pulse width modulation generation unit includes multiple pulse width modulation counters that correspond one-to-one with the physical partitions of the LED display screen. Each pulse width modulation counter is configured with an independent period register and a comparison register.

[0051] Furthermore, in another aspect of this application, in some embodiments, this application provides a content-aware adaptive control system for LED displays, referencing... Figure 4 The figure is a schematic diagram of the structure of a content-aware adaptive control system for an LED display screen according to some embodiments of this application, including: an acquisition module 201, a processing module 202, and an execution module 203, which are described below: The acquisition module 201 in this application is mainly used to acquire the video frame to be displayed, and to perform multi-scale feature extraction and channel attention recalibration on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights. Processing module 202, in this application, is used to perform semantic segmentation on the perceptual feature map through a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and to determine the semantic category label of each semantic category group; It should be noted that the processing module 202 is also used to obtain the light intensity of the current environment, and match the corresponding human eye contrast sensitivity function parameters for each semantic category group according to all semantic category labels and the preset semantic-visual mapping table, and then generate partition control parameters for different semantic category groups according to all sensitivity function parameters and the light intensity. The execution module 203 in this application is mainly used to input the partition control parameters to the driving circuit of the LED display screen for content adaptive partition display control.

[0052] The foregoing has detailed examples of the content-aware adaptive control method and system for LED displays provided in this application. It is understood that the corresponding apparatus, in order to achieve the above functions, includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specified application, but such implementation should not be considered beyond the scope of this application.

[0053] In some embodiments, this application also provides a computer device, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for calling and running the computer program from the memory, so that the computer device performs the above-described content-aware LED display adaptive control method.

[0054] In some embodiments, reference Figure 5 The dashed lines in the figure indicate that the unit or module is optional. This figure is a structural schematic diagram of a computer device implementing a content-aware adaptive control method for LED displays according to an embodiment of this application. The content-aware adaptive control method for LED displays described in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a memory 302 and at least one communication unit 305. The computer device may be a terminal device, a server or a chip.

[0055] Processor 301 can be a general-purpose processor or a special-purpose processor. For example, processor 301 can be a central processing unit (CPU), which can be used to control computer devices, execute software programs, and process data from software programs. The computer device may also include a communication unit 305 for inputting (receiving) and outputting (transmitting) signals.

[0056] For example, the computer device may be a chip, and the communication unit 305 may be the input and / or output circuit of the chip, or the communication unit 305 may be the communication interface of the chip, which may be a component of a terminal device, network device or other device.

[0057] For example, the computer device may be a terminal device or a server, and the communication unit 305 may be a transceiver of the terminal device or the server, or the communication unit 305 may be a transceiver circuit of the terminal device or the server.

[0058] The computer device may include one or more memories 302 storing a program 304. The program 304 can be executed by a processor 301 to generate instructions 303, causing the processor 301 to execute the method described in the above method embodiments according to the instructions 303. Optionally, the memory 302 may also store data (such as a target audit model). Optionally, the processor 301 may also read data stored in the memory 302, which may be stored at the same storage address as the program 304, or it may be stored at a different storage address than the program 304.

[0059] The processor 301 and memory 302 can be configured separately or integrated together, for example, integrated on the system on chip (SOC) of the terminal device.

[0060] It should be understood that each step of the above method embodiment can be completed by hardware logic circuits or software instructions in the processor 301. The processor 301 can be a CPU, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, such as discrete gates, transistor logic devices, or discrete hardware components.

[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] For example, in some embodiments, this application also provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the above-described content-aware adaptive control method for LED displays.

[0063] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0064] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A content-aware adaptive control method for LED displays, characterized in that, Includes the following steps: The video frame to be displayed is acquired, and multi-scale feature extraction and channel attention recalibration are performed on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights. The perceptual feature map is semantically segmented by a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and the semantic category label of each semantic category group is determined. The light intensity of the current environment is obtained, and the corresponding human eye contrast sensitivity function parameters are matched for each semantic category group according to all semantic category labels and the preset semantic-visual mapping table. Then, based on all the sensitivity function parameters and the light intensity, partition control parameters for different semantic category groups are generated. The zoning control parameters are input into the driving circuit of the LED display screen for content adaptive zoning display control.

2. The method as described in claim 1, characterized in that, Performing multi-scale feature extraction and channel attention recalibration on the video frame to be displayed to obtain a perceptual feature map with differentiated channel weights specifically includes: The video frame to be displayed is input into the encoder of the semantic segmentation network for multi-scale feature extraction to obtain a multi-channel feature map; The multi-channel feature map is compressed by global average pooling using a channel attention mechanism to obtain a compressed vector in the channel dimension. The compressed vector is recalibrated with channel attention to obtain a perceptual feature map with differentiated channel weights.

3. The method as described in claim 1, characterized in that, The perceptual feature map is semantically segmented using a semantic segmentation network to obtain multiple semantic category groups for pixel division in the video frame to be displayed, and the semantic category label for each semantic category group is specifically determined to include: The perceptual feature map is aggregated using a semantic segmentation network with hollow spatial pyramid pooling to obtain a feature tensor that integrates multi-scale semantic information. The feature tensor is upsampled and the low-level features of the corresponding level are fused through skip connections to obtain a semantic probability map with the same resolution as the video frame to be displayed. The pixels in the video frame to be displayed are divided into multiple semantic category groups based on the semantic category probability of each pixel in the semantic probability map. Determine the semantic category label for each semantic category group.

4. The method as described in claim 1, characterized in that, Based on all semantic category labels and a pre-defined semantic-visual mapping table, the corresponding human eye contrast sensitivity function parameters are matched for each semantic category group. Specifically, these parameters include: For each semantic category group, the pre-defined semantic-visual mapping table is queried based on the semantic category label of the semantic category group to obtain the visual saliency weight of the semantic category group; Based on the visual saliency weight, a preset human eye contrast sensitivity function model is invoked and the sensitivity curve is adjusted by weighting to obtain the human eye contrast sensitivity function parameters corresponding to the semantic category group, and then the human eye contrast sensitivity function parameters corresponding to each semantic category group are obtained.

5. The method as described in claim 1, characterized in that, Based on all the sensitivity function parameters and the illumination intensity, the specific generation of partition control parameters for different semantic category groups includes: The brightness compensation coefficient for each semantic category group is determined based on the light intensity. The sensitivity function parameters and brightness compensation coefficients of each semantic category group are weighted and fused to generate the brightness gain coefficient and gamma correction curve corresponding to each semantic category group. Based on the luminance gain coefficients and gamma correction curves of all semantic category groups, determine the partition control parameters for different semantic category groups.

6. The method as described in claim 1, characterized in that, The process of inputting the zoning control parameters into the LED display screen's driving circuit for content-adaptive zoning display control specifically includes: The zoning control parameters are transmitted to the drive circuit through the bus interface of the LED display control terminal; The driving circuit performs pixel-level brightness scaling on the video frame to be displayed according to the brightness gain coefficient of each semantic category group in the partition control parameters. The driving circuit performs a non-linear mapping on the scaled pixel values ​​in the video frame to be displayed according to the gamma correction curve of each semantic category group in the partition control parameters, and obtains the corrected driving level. The driving circuit generates a pulse width modulation signal with a corresponding duty cycle according to the driving level and applies it to the light-emitting diodes in the LED display screen corresponding to each semantic category group, so that each semantic category group presents the display content with differentiated brightness and contrast.

7. The method as described in claim 1, characterized in that, The semantic category labels include the portrait area, the text area, and the background area.

8. A content-aware adaptive control system for an LED display screen, used to execute the content-aware adaptive control method for an LED display screen as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire video frames to be displayed, and to perform multi-scale feature extraction and channel attention recalibration on the video frames to be displayed to obtain a perceptual feature map with differentiated channel weights. The processing module is used to perform semantic segmentation on the perceptual feature map through a semantic segmentation network to obtain multiple semantic category groups of pixels in the video frame to be displayed, and to determine the semantic category label of each semantic category group. The processing module is also used to obtain the light intensity of the current environment, and match the corresponding human eye contrast sensitivity function parameters for each semantic category group according to all semantic category labels and the preset semantic-visual mapping table, and then generate partition control parameters for different semantic category groups according to all sensitivity function parameters and the light intensity. The execution module is used to input the partition control parameters to the driving circuit of the LED display screen for content adaptive partition display control.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the content-aware adaptive control method for an LED display screen as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the content-aware adaptive control method for LED displays as described in any one of claims 1 to 7.

Images