A method and system for controlling the brightness of an LCD display

Abstract

The present application relates to display control technical field, specifically to a kind of LCD display brightness control method and system, method is applied to the display device containing direct backlight module and liquid crystal panel, backlight module is the multi-partition light source of independently adjustable brightness, method includes: constructing backlight module diffusion simulation module, based on point spread function to backlight brightness distribution diffusion obtains backlight output image;Utilize training sample to train HDR reconstruction model and local dimming model, HDR reconstruction model is based on saturation discrimination generation feature mask and is propagated in convolution feature extraction to inhibit saturated area feature, local dimming model generates supervision signal by means of diffusion simulation module and is updated with the loss function containing image fidelity term and power consumption constraint term;Display stage carries out luminance correction to input low dynamic range image and combines feature mask and exports target HDR image;Again output backlight prediction image and obtain backlight output image by diffusion simulation;According to the partition topological partition convergence forms backlight control image, and by target HDR image and backlight output image generates liquid crystal panel control image representation pixel transmittance;Accordingly drive each partition luminous intensity and each pixel transmittance, realize backlight and liquid crystal collaborative modulation output display image.The present application generates backlight partition control and liquid crystal panel transmittance control, so that display output approximates target HDR image and gives consideration to power consumption constraint and halo artifact suppression.

Description

Technical Field

[0001] This invention relates to the field of display control technology, and in particular to a method and system for controlling the brightness of an LCD display. Background Technology

[0002] Current LCDs typically employ constant, uniform global backlighting. When displaying black levels, light leakage from the liquid crystal can result in insufficient black depth, thus lowering contrast. Furthermore, continuous backlight illumination leads to additional power consumption. Direct-lit (multi-zone) backlighting with local dimming is considered an effective way to improve contrast and energy efficiency. However, achieving coordinated modulation of backlight zone emission and liquid crystal pixel transmittance often requires paired supervision data for both the backlight and the panel during the training phase. In the absence of readily available hardware sampling and public training sets, traditional supervised learning is difficult to implement. Therefore, it is necessary to establish a backlight diffusion simulation consistent with the diffuser (such as a diffusion model based on the point spread function PSF) to generate usable training supervision signals. In addition, single-image LDR to HDR reconstruction encounters problems such as unreliable information in saturated highlight areas and the introduction of artifacts. It is necessary to suppress the influence of invalid features in saturated areas during feature extraction. Summary of the Invention

[0003] In view of the above technical problems, the present invention provides an LCD display brightness control method and system, which aims to solve the problem of how to robustly reconstruct the target HDR brightness distribution in a direct-lit multi-zone backlit LCD under the condition of only inputting a single LDR image, and how to construct a trainable supervision signal through backlight diffusion simulation under the conditions of lack of real backlight output samples / public training sets and lack of hardware sampling, so as to jointly generate backlight zone control and liquid crystal panel transmittance control, so that the display output approximates the target HDR image while taking into account power consumption constraints and halo artifact suppression.

[0004] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part by practice of this disclosure.

[0005] According to one aspect of the present invention, an LCD display brightness control method is provided, the method being applied to a display device including a direct-lit backlight module and a liquid crystal panel, wherein the direct-lit backlight module has a multi-zone light source with independently adjustable brightness, characterized in that the method includes: A backlight module diffusion simulation module is constructed. The backlight module diffusion simulation module performs diffusion operation on the backlight brightness distribution based on the point diffusion function to output the backlight output image. The HDR reconstruction model and the local dimming model are trained using training samples. When training the HDR reconstruction model, a feature mask is generated based on the saturation discrimination condition and the feature mask is propagated during the convolution feature extraction process to suppress saturated region features. When training the local dimming model, a supervision signal is generated by the backlight module diffusion simulation module in the absence of physical backlight output samples, and the model parameters are updated using a loss function that includes image fidelity terms and power consumption constraints. During the display stage, an input low dynamic range image is acquired, the brightness of the input low dynamic range image is corrected, and the brightness-corrected image and the feature mask are input into the trained HDR reconstruction model to output the target HDR image. The target HDR image is input into the trained local dimming model to output a backlight prediction image, and the backlight prediction image is input into the backlight module diffusion simulation module to obtain the backlight output image; The backlight output image is partitioned and converged according to the partition topology of the direct-lit backlight module to obtain a backlight control image, and a liquid crystal panel control image is generated according to the target HDR image and the backlight output image. The liquid crystal panel control image represents the pixel transmittance of the liquid crystal panel. The backlight control image drives the luminous intensity of each zone of the direct-lit backlight module, and the liquid crystal panel control image drives the transmittance of each pixel of the liquid crystal panel, so that the display device outputs a display image corresponding to the target HDR image in a backlight and liquid crystal co-modulation mode.

[0006] Furthermore, the generation of the feature mask based on the saturation discrimination condition includes: The pixel brightness of the input low dynamic range image is normalized and compared with a preset saturation threshold; when the pixel brightness is lower than the preset saturation threshold, a larger mask weight is set for the corresponding pixel. When the pixel brightness reaches or exceeds the preset saturation threshold, the mask weight is monotonically compressed according to the degree to which the pixel brightness approaches the maximum brightness, so that the pixel with higher saturation corresponds to a smaller mask weight, thereby suppressing its influence on feature extraction without completely discarding the information of the saturated region. The brightness correction includes performing nonlinear gamma correction on the input low dynamic range image to adjust the brightness distribution; The output target HDR image includes: using the feature mask as weight, performing pixel-by-pixel weighted fusion of the brightness-corrected image and the predicted HDR components output by the HDR reconstruction model, so that the fusion result of the normally exposed area mainly comes from the brightness-corrected image, and the fusion result of the saturated area mainly comes from the predicted HDR components.

[0007] Furthermore, the HDR reconstruction model performs feature masking operations at each convolutional layer, including: multiplying the feature map of the convolutional layer element-wise with the corresponding layer feature mask to obtain a mask feature map, and then inputting the mask feature map into the convolutional layer to generate the next layer feature map; and constructing a normalized weight kernel based on the absolute value of the convolutional kernel weights of the convolutional layer, and performing convolution propagation of the normalized weight kernel with the corresponding layer feature mask to obtain the next layer feature mask, wherein a preset stabilizing term is introduced during the normalization process.

[0008] Furthermore, the HDR reconstruction model includes a U-shaped structure composed of an encoder and a decoder. The encoder extracts multi-scale features through step-by-step downsampling, and the decoder restores spatial resolution through step-by-step upsampling, setting skip connections between corresponding scales to fuse encoded and decoded features. The convolutional layers of the encoder use a rectified activation function with leakage or an equivalent nonlinear activation function, the convolutional layers of the decoder use a rectified activation function, and the output layer uses linear activation to output continuous HDR values. The upsampling uses nearest neighbor interpolation, bilinear interpolation, or a combination thereof, and the feature mask operation is integrated in at least some of the convolutional layers of the encoder and the decoder.

[0009] Furthermore, the loss function used to train the HDR reconstruction model includes: HDR reconstruction loss is used to calculate the aggregate amount of the absolute value of the pixel difference between the saturated region of the target HDR image and the corresponding region of the reference HDR image in the logarithmic domain, wherein the saturated region is selected to participate in the calculation through the complementary weights of the feature mask; The perceptual loss includes feature loss and style loss. The feature loss is obtained by compressing the target HDR image and the reference HDR image into a pre-trained feature extraction network to obtain multi-layer feature maps, and then aggregating the differences between the corresponding layer feature maps. The style loss is obtained by calculating Gram matrices reflecting channel correlation for each of the multi-layer feature maps, and then aggregating the differences between the corresponding layer Gram matrices. Specifically, the HDR reconstruction loss and the perception loss are combined with preset weights to update the HDR reconstruction model parameters.

[0010] Furthermore, the local dimming model is a segmentation network with an encoder-decoder structure or an improved network thereof. The encoder includes a multi-level convolutional module composed of residual connections to enhance gradient propagation. The decoder includes a step-by-step upsampling module and establishes cross-layer connections with the corresponding level features of the encoder through concatenation or addition. The input of the local dimming model is obtained by combining the color channels of the target HDR image, the brightness channel calculated from the target HDR image, and the ideal backlight image. The output of the local dimming model is a backlight prediction image with the same spatial resolution as the target HDR image, and the backlight prediction value is limited to a driveable brightness range at the output end through a bounded mapping function.

[0011] Furthermore, the backlight module diffusion simulation module includes: A preset point diffusion function kernel is used to characterize the light diffusion properties of the diffuser plate; The backlight prediction image is convolved with the point spread function kernel to obtain the backlight output image, and mirror filling, copy filling or periodic filling are used at the convolution boundary to reduce boundary artifacts. The partition aggregation includes: dividing the backlight output image into multiple regions corresponding to the partitions based on the number and location of the partitions of the direct-lit backlight module, and performing averaging, weighted averaging or statistical robust aggregation on the pixel brightness of each region to generate control values ​​for the corresponding partitions, thereby forming the backlight control image.

[0012] Furthermore, the generation of the liquid crystal panel control image includes: Using the target HDR image as the desired display brightness distribution and the backlight output image as the incident backlight brightness distribution, the transmittance is obtained by solving pixel by pixel, so that the result of multiplying the transmittance by the backlight output image pixel by pixel approximates the target HDR image. The obtained transmittance is clipped or bounded mapping to meet the transmittance range constraints that the liquid crystal pixel can achieve; in color processing, the transmittance is solved separately on multiple color channels or uniformly on the luminance channel and combined with color ratio recovery to reduce color distortion and maintain consistency between highlight and shadow details.

[0013] Furthermore, the image fidelity item includes: The difference between the simulated HDR image and the target HDR image in the pixel domain or the perceptual domain is measured to obtain a regression loss. The regression loss is used to drive the backlight prediction image output by the local dimming model and the liquid crystal panel control image to reproduce the target HDR image under co-modulation. The power consumption constraint includes: The overall brightness level of the backlight prediction image is aggregated and added as a penalty term to the loss function to suppress excessively bright backlight output and reduce backlight energy consumption.

[0014] According to another aspect of the present invention, an LCD display brightness control system is provided, comprising: A module is used to construct a backlight module diffusion simulation module. The backlight module diffusion simulation module performs diffusion operations on the backlight brightness distribution based on the point diffusion function to output a backlight output image. The model training module is used to train the HDR reconstruction model and the local dimming model using training samples. When training the HDR reconstruction model, a feature mask is generated based on the saturation discrimination condition and the feature mask is propagated during the convolution feature extraction process to suppress saturated region features. When training the local dimming model, a supervision signal is generated by the backlight module diffusion simulation module in the absence of physical backlight output samples, and the model parameters are updated using a loss function that includes image fidelity terms and power consumption constraints. The correction module is used to acquire the input low dynamic range image during the display stage, perform brightness correction on the input low dynamic range image, and input the brightness-corrected image and the feature mask into the trained HDR reconstruction model to output the target HDR image. The prediction module is used to input the target HDR image into the trained local dimming model to output a backlight prediction image, and input the backlight prediction image into the backlight module diffusion simulation module to obtain the backlight output image; A brightness control module is used to partition and converge the backlight output image according to the partition topology of the direct-lit backlight module to obtain a backlight control image, and generate a liquid crystal panel control image according to the target HDR image and the backlight output image, wherein the liquid crystal panel control image represents the pixel transmittance of the liquid crystal panel. The display control module is used to drive the luminous intensity of each zone of the direct-lit backlight module according to the backlight control image, and to drive the transmittance of each pixel of the liquid crystal panel according to the liquid crystal panel control image, so that the display device outputs a display image corresponding to the target HDR image in a backlight and liquid crystal co-modulation mode.

[0015] The technical solution of the present invention has the following beneficial effects: Compared to data-driven solutions that rely on empirical rules or require actual hardware sampling, this invention uses a software simulator for the backlight module to match the diffuser function, reducing hardware modification / sampling costs and enabling training even in the absence of training datasets. On the HDR reconstruction side, a saturation-based feature mask is introduced and propagated between network layers to suppress invalid features in saturated regions. This allows the reconstructed HDR to be a weighted fusion of normally exposed areas and predicted HDR components, reducing saturation artifacts and improving detail and visual consistency. On the local dimming side, PSF diffusion modeling and partition aggregation are combined to more closely approximate the actual backlight diffusion output. Optimization alleviates common halo problems while improving contrast and reducing overall power consumption, achieving an HDR display effect through backlight and liquid crystal co-modulation. Attached Figure Description

[0016] Figure 1 This is a flowchart of an LCD display brightness control method according to an embodiment of this specification; Figure 2 This is a structural block diagram of an LCD display brightness control system according to an embodiment of this specification; Figure 3 This specification describes a terminal device that implements an LCD display brightness control method in one of its embodiments. Figure 4 This specification describes a computer-readable storage medium storing an LCD display brightness control method as an example of an embodiment of the present specification. Detailed Implementation

[0017] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, systems, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0018] Furthermore, the accompanying drawings are merely illustrative of this disclosure. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0019] This invention provides a method for controlling the brightness of an LCD display. (Refer to...) Figure 1 The diagram shown is a flowchart illustrating an LCD display brightness control method according to an embodiment of the present invention. This method can be applied to display devices including a direct-lit backlight module and a liquid crystal panel. The direct-lit backlight module has multi-zone light sources with independently adjustable brightness. Specifically, the method may include the following steps S101-S106: In step S101, a backlight module diffusion simulation module is constructed. The backlight module diffusion simulation module performs diffusion operation on the backlight brightness distribution based on the point diffusion function to output the backlight output image.

[0020] In order to approximate the light output distribution of a direct-lit backlight module after passing through a diffuser plate on the software side, a backlight module diffusion simulation module is constructed. The diffusion effect of the diffuser plate on the light source is characterized by a point spread function kernel g, and the backlight brightness distribution image is used as input to obtain the backlight output image through discrete convolution; where the input backlight brightness distribution can be denoted as... The output backlight image can be denoted as The diffusion operation satisfies:

[0021] In the formula, and Let represent the width and height of the point spread function filter, respectively. This represents the diffusion weight of the diffuser plate at a spatial offset (x,y). This simulation module can make the input backlight brightness distribution in space produce the same diffusion and superposition effect as the diffuser plate without relying on physical backlight sampling, thereby obtaining the backlight output image for subsequent processing.

[0022] In step S102, the HDR reconstruction model and the local dimming model are trained using training samples. When training the HDR reconstruction model, a feature mask is generated based on the saturation discrimination condition and the feature mask is propagated during the convolution feature extraction process to suppress saturated region features. When training the local dimming model, a supervision signal is generated by the backlight module diffusion simulation module in the absence of physical backlight output samples, and the model parameters are updated using a loss function that includes image fidelity terms and power consumption constraints.

[0023] The step of generating a feature mask based on saturation discrimination conditions includes: normalizing the pixel brightness of the input low dynamic range image and comparing it with a preset saturation threshold; setting a larger mask weight for the corresponding pixel when the pixel brightness is lower than the preset saturation threshold; and monotonically compressing the mask weight according to the degree to which the pixel brightness approaches the maximum brightness when the pixel brightness reaches or exceeds the preset saturation threshold, so that pixels with higher saturation correspond to smaller mask weights, thereby suppressing their influence on feature extraction without completely discarding saturated region information.

[0024] The HDR reconstruction model performs feature masking operations at each convolutional layer, including: multiplying the feature map of the convolutional layer element-wise with the feature mask of the corresponding layer to obtain a mask feature map, and then inputting the mask feature map into the convolutional layer to generate the feature map of the next layer; and constructing a normalized weight kernel based on the absolute value of the weights of the convolutional kernel of the convolutional layer, and performing convolution propagation of the normalized weight kernel with the feature mask of the corresponding layer to obtain the feature mask of the next layer, wherein a preset stabilizing term is introduced in the normalization process.

[0025] The HDR reconstruction model includes a U-shaped structure consisting of an encoder and a decoder. The encoder extracts multi-scale features through stepwise downsampling, and the decoder restores spatial resolution through stepwise upsampling. Skip connections are set between corresponding scales to fuse encoded and decoded features. The convolutional layers of the encoder use a rectified activation function with leakage or an equivalent nonlinear activation function, the convolutional layers of the decoder use a rectified activation function, and the output layer uses linear activation to output continuous HDR values. The upsampling uses neighbor interpolation, bilinear interpolation, or a combination thereof, and the feature mask operation is integrated in at least some of the convolutional layers of the encoder and the decoder.

[0026] The loss function used to train the HDR reconstruction model includes: HDR reconstruction loss, which is used to calculate the aggregated absolute value of the pixel difference between the saturated region of the target HDR image and the corresponding region of the reference HDR image in the logarithmic domain, wherein the saturated region is selected to participate in the calculation through the complementary weights of the feature mask; perceptual loss, including feature loss and style loss, wherein the feature loss is obtained by inputting the target HDR image and the reference HDR image into a pre-trained feature extraction network after numerical range compression to obtain multi-layer feature maps, and aggregating the differences of the corresponding layer feature maps; the style loss is obtained by calculating the Gram matrix reflecting the channel correlation for the multi-layer feature maps respectively, and aggregating the differences of the corresponding layer Gram matrix; wherein the HDR reconstruction loss and the perceptual loss are combined according to preset weights to update the parameters of the HDR reconstruction model.

[0027] The image fidelity term includes: measuring the difference between the simulated HDR image and the target HDR image in the pixel domain or perceptual domain to obtain a regression loss, which is used to drive the backlight prediction image output by the local dimming model and the liquid crystal panel control image to reproduce the target HDR image under co-modulation.

[0028] The power consumption constraint includes: performing an aggregate calculation on the overall brightness level of the backlight prediction image and adding it as a penalty term to the loss function to suppress excessively bright backlight output and reduce backlight power consumption.

[0029] As an explanation, training samples are selected from pairs of low dynamic range (LVR) and high dynamic range (HDR) image data. Supervised learning is used to update the parameters of the HDR reconstruction network, enabling the network to recover information lost in saturated regions from a single LVR input and mitigating checkerboard or halo artifacts caused by ambiguity in convolution on saturated pixels. Simultaneously, it provides a high dynamic range input foundation for subsequent local dimming networks. During training, the pixel brightness of the input LVR image is normalized to [0,1] and a saturation discrimination threshold t is introduced. When the brightness is below the threshold, the mask weights are reset to 1. When the brightness reaches or exceeds the threshold, the mask weights are compressed based on how close the brightness is to the maximum value, so that pixels closer to the maximum brightness correspond to smaller weights. Specifically, this is defined as:

[0030] Where S is the pixel brightness value, t is a preset threshold, and the mask value is limited to [0,1]. Weak weights in saturated regions are not completely discarded but participate in feature propagation with a lower contribution, thus suppressing invalid content while retaining usable clues. During training, the mask is embedded into the convolutional feature extraction chain and propagated layer by layer: the k-th layer feature map... With corresponding mask Element-wise multiplication yields the mask feature map. Their relationship is ; in For pixel-wise multiplication, a mask value of 1 indicates that the feature is considered reasonably exposed, and a mask value of 0 indicates that the feature is considered invalid. Then, the output of the k-th layer is processed by convolution, bias, and non-linearity to obtain the features for the next layer. ; in and These represent the convolution weights and biases for this layer, respectively; * indicates the standard convolution operation. The activation function is used; the mask propagation is achieved by normalizing the absolute value of the convolution kernel and then convolving it with the previous layer mask, ensuring that the mask expansion is consistent with the receptive field expansion between layers. ; in The element-wise absolute value of the weight tensor. For L1 norm normalization term, To prevent the stability term from division by zero, the network structure adopts a U-shaped encoder-decoder framework. The encoder downsamples step by step to extract multi-scale features, and the decoder upsamples step by step to restore spatial resolution. Information from different scales is fused through cross-layer connections. The above masking operation is performed in the convolutional layers, and upsampling can be performed using methods such as nearest neighbor interpolation to cooperate with the feature restoration at the decoder.

[0031] To ensure that HDR reconstruction focuses more on detail recovery in saturated areas while also maintaining perceptual quality, the training loss is a weighted average of the HDR reconstruction loss and the perceptual loss: ; The HDR reconstruction loss measures the L1 difference between the output image and the corresponding region of the reference high dynamic range image in the logarithmic domain, and selects the saturated region for calculation using complementary weights (1-M).

[0032] in H represents the high dynamic range component predicted by the network, and H is the reference high dynamic range image. The perceptual loss consists of deep feature differences and style differences. ; Deep feature loss extracts multi-layer features and calculates differences by applying a numerical range compression function R to the input and then feeding it into VGG19. ; in This is the feature map of the k-th layer of VGG19. To learn more realistic textures and appearances in saturated regions, a fused image is introduced. Combine the normally exposed content of the reference image with the saturated region content of the predicted result: ; Style loss is measured by the Gram matrix to account for differences in channel correlation. ; in The normalization constant is The size is Training data can come from public datasets containing pairs of low and high dynamic range datasets; optimization uses Adam and sets the initial learning rate. The weights decay to 0.5 every 20 epochs, with a batch size of 16 and a total of 200 training epochs. Robustness is enhanced through random pruning, horizontal flipping, and brightness perturbation. The threshold t can be set to 0.8, and the loss weights... to The regularization coefficient and other parameters are set empirically.

[0033] The local dimming network is trained using high dynamic range (HDR) images as input, aiming to directly predict the output intensity distribution of backlight LEDs. In the absence of actual display hardware and corresponding backlight output samples, a software-based matching simulation of the diffusion process is used to establish a supervised display link, enabling training to rely on HDR datasets. The network uses UNet as its backbone and takes HDR RGB images as input. With ideal backlight image In the formation of channel dimension combination The input consists of an ideal backlight image where each pixel is assigned a usable light source in a pixel-to-pixel correspondence to characterize the desired backlight; the network output is a full-resolution single-channel backlight prediction. And at the end, it is limited to a driveable range by a sigmoid function. To obtain simulated display results that can be modulated in conjunction with the liquid crystal transmittance, the backlight prediction is diffused and the diffused backlight incident distribution is obtained through diffusion simulation. And compared with the transmittance diagram of the LCD panel. Pixel-by-pixel multiplication creates a simulated high dynamic range image: ; Based on this, a loss update network parameter is constructed that includes image fidelity terms and power consumption constraints. The total loss is written as: ; in It is a regression-type fidelity loss used to measure the difference between the simulated display result and the target high dynamic range image in the pixel domain or perceptual domain to drive reproduction consistency. The brightness regularization term, which aggregates the overall brightness level of the predicted backlight and uses it as a penalty to suppress excessive backlight brightness in order to reduce energy consumption, is defined as follows: ; in This represents the maximum output value of the backlight module. This method is used to balance the gradient contributions of the two parts to improve convergence. Through this training method, the local dimming network can still obtain end-to-end supervision signals by means of diffusion simulation and display links even without physical backlight output samples, and avoids the checkerboard halo problem that is easy to produce by the traditional empirical thresholding method.

[0034] In step S103, an input low dynamic range image is acquired during the display stage, the brightness of the input low dynamic range image is corrected, and the brightness-corrected image and the feature mask are input into the trained HDR reconstruction model to output the target HDR image.

[0035] The brightness correction includes performing nonlinear gamma correction on the input low dynamic range image to adjust the brightness distribution. The output target HDR image includes: using the feature mask as weight, performing pixel-wise weighted fusion of the brightness-corrected image and the predicted HDR components output by the HDR reconstruction model, so that the fusion result of the normally exposed areas mainly comes from the brightness-corrected image, and the fusion result of the saturated areas mainly comes from the predicted HDR components.

[0036] Specifically, in step S103, the display stage receives a low dynamic range image as input, normalizes the pixel value range to [0,1], and performs nonlinear adjustment on the brightness distribution based on the idea of ​​gamma correction, obtaining the brightness-corrected image through power-law transformation. ,in It is a gamma correction factor and satisfies This is used to redistribute the brightness of input pixels to improve the available information density in dark and mid-brightness regions. The brightness-corrected image and feature mask M are then fed into a trained HDR reconstruction model, which outputs predicted HDR components. Then, using the mask as a pixel-wise weight, a weighted fusion is performed between the input correction result and the predicted HDR components to obtain the target HDR image. Its expression is: ; in Represents pixel-wise product, mask Generated from the network, M and 1-M respectively characterize the contributions of the normally exposed region and the saturated region in the fusion process. The first term on the right corresponds to the normally exposed region, which is mainly composed of... Provided, the second item on the right corresponds to the saturated region mainly composed of This method provides a way to fill in the missing details in saturated areas with network predictions while maintaining the original structure and color trends of normal areas. This fusion method ensures that saturated pixels do not enter the final image with the same weight as normal pixels. The weak mask value still allows limited cues from saturated areas to participate in the reconstruction, avoiding abrupt boundary changes caused by complete discarding. Simultaneously, it suppresses interference from saturated areas in the reconstruction and improves the visibility of highlight details.

[0037] In step S104, the target HDR image is input into the trained local dimming model to output a backlight prediction image, and the backlight prediction image is input into the backlight module diffusion simulation module to obtain the backlight output image.

[0038] Specifically, in step S104, the target HDR image obtained in step S103 is used as the input to the local dimming model to predict the backlight brightness distribution. The local dimming model employs an encoder-decoder network to output a backlight prediction image at full resolution. The input is combined in the channel dimension, including both the color information of the target HDR image and the brightness information calculated from the target HDR image. An ideal backlight image is introduced as an auxiliary prior, enabling the network to simultaneously consider the correspondence between color, brightness, and the desired backlight shape when learning the backlight distribution. The network output is a backlight prediction image with the same spatial size as the target HDR image. The output end uses a bounded mapping function to limit the predicted value to the driveable brightness range, thereby avoiding unattainable negative values ​​or out-of-range brightness, and enabling the prediction results to be directly used for backlight driving. After obtaining the backlight prediction image, it is input into the backlight module diffusion simulation module to form the backlight output image under the diffusion plate effect. This process corresponds to checking with the point spread function. Diffusion calculations are performed to reflect the spatial superposition and energy diffusion of the direct-lit multi-zone light source after passing through the diffuser plate, making the incident backlight distribution used in subsequent backlight-liquid crystal co-modulation closer to the actual light output pattern. The diffused backlight output image exhibits a smoother transition in space, which can convert the fine-grained backlight changes output by the local dimming network into the actual incident distribution after the diffuser plate, providing consistent backlight input conditions for subsequent zone convergence and liquid crystal transmittance calculation.

[0039] In step S105, the backlight output image is partitioned and converged according to the partition topology of the direct-lit backlight module to obtain a backlight control image, and a liquid crystal panel control image is generated according to the target HDR image and the backlight output image. The liquid crystal panel control image represents the pixel transmittance of the liquid crystal panel.

[0040] The process of generating the control image for the liquid crystal panel includes: using the target HDR image as the desired display brightness distribution and the backlight output image as the incident backlight brightness distribution, obtaining the transmittance based on a pixel-by-pixel solution, and making the result of multiplying the transmittance by the backlight output image pixel by pixel approximate the target HDR image; performing clipping or bounded mapping on the obtained transmittance to meet the transmittance range constraints that the liquid crystal pixels can achieve; in color processing, the transmittance is solved separately on multiple color channels or uniformly on the brightness channel and combined with color ratio recovery to reduce color distortion and maintain consistency between highlight and shadow details.

[0041] In the display stage, after obtaining the diffused backlight output image, this image can be considered as the backlight brightness distribution emitted from the diffusion layer onto the incident surface of the liquid crystal panel. The diffusion process can be characterized by performing convolution on the backlight prediction image using a point spread function. The resulting backlight brightness distribution is denoted as... During zone convergence, to match the number and spatial arrangement of LED zones in the direct-lit backlight module, The backlight is evenly divided into multiple rectangular blocks according to the partition location. Each block corresponds to an independently driveable partition, and the average brightness of all pixels in the block is used to replace all pixels in that block, thus forming a partition-level backlight control image to meet the requirement of driving the actual partition brightness with partition control values; the LCD panel control image is used to characterize the transmittance distribution of each pixel. The coordinated modulation relationship between the displayed image, backlight, and liquid crystal can be expressed as: ,in For pixel-by-pixel multiplication, To simulate and display HDR images after co-modulation, Driven by the grayscale of liquid crystal pixels and reflecting the proportion of light transmission, this product relationship can be solved pixel by pixel. Approximating the desired display brightness distribution In practice, the transmittance obtained by the solution can be clipped or bounded to meet the transmittance range constraints that the liquid crystal can achieve. In color processing, the transmittance can be solved separately for each color channel, or the transmittance can be solved uniformly for the brightness channel and then combined with the color ratio recovery to reduce color distortion and maintain the consistency of highlights and shadow details.

[0042] In step S106, the luminous intensity of each zone of the direct-lit backlight module is driven according to the backlight control image, and the transmittance of each pixel of the liquid crystal panel is driven according to the liquid crystal panel control image, so that the display device outputs a display image corresponding to the target HDR image in a backlight and liquid crystal co-modulation mode.

[0043] Specifically, in step S106, the backlight control image is used as the driving basis for the direct-lit backlight module. The control value corresponding to each zone is mapped to the driving intensity of the light source in that zone. The driving intensity can be achieved through current amplitude, PWM duty cycle, or a combination of both, so that the output light emission level of the zone light source is consistent with the control value. The liquid crystal panel control image is used as the pixel-level transmittance control basis. The transmittance value of each pixel is mapped to the grayscale of the liquid crystal pixel or the equivalent driving amount, so that the light transmittance ratio of the pixel to the incident backlight meets the requirements of the control image. The coordinated modulation of the backlight and the liquid crystal satisfies the pixel-by-pixel product relationship, and the simulated HDR result on the display side is as described above. .

[0044] When the backlight intensity and pixel transmittance are updated synchronously, the product result spatially approximates the brightness distribution of the target HDR image, so that the bright areas are achieved by higher backlight and higher transmittance, and the dark areas are achieved by lower backlight and lower transmittance. This maintains the level of detail in both bright and dark areas while avoiding the black level rise caused by global backlighting, and limits the backlight output within a controllable range to take into account energy consumption constraints.

[0045] In one embodiment, the local dimming model is a segmentation network with an encoder-decoder structure or an improved network thereof. The encoder includes a multi-level convolutional module composed of residual connections to enhance gradient propagation. The decoder includes a step-by-step upsampling module and establishes cross-layer connections with the corresponding level features of the encoder through concatenation or addition. The input of the local dimming model is obtained by combining the color channels of the target HDR image, the brightness channel calculated from the target HDR image, and the ideal backlight image. The output of the local dimming model is a backlight prediction image with the same spatial resolution as the target HDR image, and the backlight prediction value is limited to a driveable brightness range at the output end through a bounded mapping function.

[0046] The local dimming model employs an encoder-decoder structure, using a segmented network to predict backlight intensity at each spatial location. The network itself consists of multiple convolutional layers. The encoder downsamples and extracts features at each resolution until a low-resolution threshold is reached, then proceeds to the decoder for upsampling to restore the resolution. At each corresponding resolution, the encoder features are directly passed to the decoder. Cross-layer connections fuse multi-scale information to obtain a larger receptive field and accelerate optimization convergence. The encoder can incorporate multi-level convolutional modules composed of residual connections. The residual blocks add the computational output of each block to the input, making gradients easier to propagate and improving the training stability of deep networks. One implementation uses an 18-layer convolutional residual network as the backbone. Except for the first layer, which uses a 7×7 kernel with a stride of 1, the remaining convolutional layers use 3×3 kernels. The decoder can contain multiple upsampling layers. One implementation uses bilinear upsampling, followed by deconvolution kernels, normalization, and activation functions. Simultaneously, it matches and connects with features from the encoder at the corresponding layer, thus preserving structural information at the encoder while restoring spatial details. The activation function can be ReLU, combined with normalization to aid convergence optimization. The input is combined in the channel dimension, allowing the network to simultaneously obtain color and luminance cues as well as an ideal backlight prior: the RGB channels of the target HDR image are concatenated with the luminance channel calculated from the target HDR image, and the ideal backlight image is used as an additional channel in the combination, forming a multi-channel input tensor. In one implementation, the input HDR RGB image and the ideal backlight image are combined to obtain a tensor of size [size missing]. The input is defined by a pixel-to-pixel correspondence, where each pixel has available LED elements as a priori light source to guide the learning of the backlight distribution. The output is a single-channel backlight prediction image with the same spatial resolution as the target HDR image, and its numerical range is limited to [0,1]. Bounded mapping can be achieved by adding a sigmoid function after the final convolution, thereby constraining the predicted value within the driveable brightness range, making it easy to use directly for backlight control.

[0047] In one embodiment, the backlight module diffusion simulation module includes: a preset point spread function kernel for characterizing the light diffusion characteristics of the diffuser plate; performing a convolution operation on the backlight prediction image and the point spread function kernel to obtain the backlight output image, and using mirror filling, copy filling or periodic filling at the convolution boundary to reduce boundary artifacts; the partition aggregation includes: dividing the backlight output image into multiple regions corresponding to the partitions according to the number and position of the partitions of the direct-lit backlight module, and performing averaging, weighted averaging or statistical robust aggregation on the pixel brightness in each region to generate control values ​​for the corresponding partitions, thereby forming the backlight control image.

[0048] In one embodiment, the backlight module diffusion simulation module is pre-defined with a point diffusion function kernel g to characterize the spatial diffusion characteristics of the diffuser plate, and uses the backlight prediction image. As input, the backlight output image is obtained by simulating the diffusion and superposition of light energy by a diffuser plate through discrete convolution. Their relationship satisfies:

[0049] in and Here, g(x,y) represents the width and height of the point spread function filter, respectively. g(x,y) reflects the weight of the diffuser's energy contribution at different spatial offsets, and the convolution result corresponds to the backlight brightness distribution behind the diffuser. To reduce brightness breaks or stripe artifacts at convolution boundaries due to a lack of neighborhood information, the convolution boundary can be expanded using mirror padding, copy padding, or periodic padding before performing the convolution operation, making the calculation of edge regions closer to the statistical characteristics of the image's interior. To ensure the backlight output matches the partition topology of the direct-lit backlight module, partition convergence evenly divides the backlight output image into multiple rectangular regions based on the number and spatial location of LED partitions. Each region corresponds to one partition, and the pixel brightness within each region is replaced with the average value to generate partition-level control values, forming a backlight control image. This ensures a one-to-one correspondence between the backlight control quantities and the actual partition drives. In addition to arithmetic mean aggregation within a partition, weighted average can be performed by setting weights based on the distance from the partition center to the pixel to emphasize the contribution of the dominant region of the partition, or statistically robust aggregation that is insensitive to outliers can be used to suppress the pull of local peak brightness on the partition control value. This reduces the halo tendency caused by abrupt changes at the partition boundary under the condition of a limited number of partitions, and maintains the spatial continuity and driveability of the backlight control image.

[0050] Based on the same line of thought, such as Figure 2 As shown, an LCD display brightness control system is provided, comprising: Module 201 is used to construct a backlight module diffusion simulation module. The backlight module diffusion simulation module performs diffusion operations on the backlight brightness distribution based on the point diffusion function to output a backlight output image. The model training module 202 is used to train the HDR reconstruction model and the local dimming model using training samples. When training the HDR reconstruction model, a feature mask is generated based on the saturation discrimination condition and the feature mask is propagated during the convolution feature extraction process to suppress saturated region features. When training the local dimming model, a supervision signal is generated by the backlight module diffusion simulation module in the absence of physical backlight output samples, and the model parameters are updated using a loss function that includes image fidelity terms and power consumption constraints. The correction module 203 is used to acquire an input low dynamic range image during the display stage, perform brightness correction on the input low dynamic range image, and input the brightness-corrected image and the feature mask into the trained HDR reconstruction model to output a target HDR image. Prediction module 204 is used to input the target HDR image into the trained local dimming model to output a backlight prediction image, and input the backlight prediction image into the backlight module diffusion simulation module to obtain the backlight output image; The brightness control module 205 is used to partition and converge the backlight output image according to the partition topology of the direct-lit backlight module to obtain a backlight control image, and generate a liquid crystal panel control image according to the target HDR image and the backlight output image. The liquid crystal panel control image represents the pixel transmittance of the liquid crystal panel. The display control module 206 is used to drive the luminous intensity of each zone of the direct-lit backlight module according to the backlight control image, and drive the transmittance of each pixel of the liquid crystal panel according to the liquid crystal panel control image, so that the display device outputs a display image corresponding to the target HDR image in a backlight and liquid crystal co-modulation mode.

[0051] Compared to data-driven solutions that rely on empirical rules or require actual hardware sampling, this system reduces hardware modification / sampling costs by constructing a software simulator for the backlight module to match the diffuser plate function, and can still conduct training even when training datasets are lacking. On the HDR reconstruction side, a feature mask based on saturation discrimination is introduced and propagated between network layers to suppress invalid features in saturated regions. This allows the reconstructed HDR to be a weighted fusion of the normal exposure area and the predicted HDR component, thereby reducing saturation artifacts and improving detail and visual consistency. On the local dimming side, PSF diffusion modeling and partition aggregation are combined to more closely approximate the real backlight diffusion output. Through optimization, common halo problems are alleviated, while improving contrast and reducing overall power consumption, achieving an HDR display effect with backlight and liquid crystal co-modulation.

[0052] The specific details of each module / unit in the above system have been described in detail in the implementation method section. For any undisclosed details, please refer to the implementation method section, and therefore will not be repeated here.

[0053] Following the same line of thought, this specification also provides an LCD display brightness control device, such as... Figure 3 As shown.

[0054] The LCD display brightness control device can be the terminal device or server provided in the above embodiments.

[0055] LCD display brightness control devices can vary significantly depending on configuration or performance. They may include one or more processors 301, memory 302, and buses. Memory 302 may store one or more application programs or data. Memory 302 may include readable media in the form of volatile memory cells, such as random access memory (RAM) and / or cache memory cells, i.e., plug-in external hard drives, smart media cards (SMC), secure digital cards (SD cards), flash cards, etc., and may further include read-only memory cells. The application programs stored in memory 302 may include one or more program modules (not shown in the figures). Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Furthermore, processor 301 may be configured to communicate with memory 302 and execute a series of computer-executable instructions stored in memory 302 on the LCD display brightness control device. The LCD display brightness control device may also include one or more power supplies 303, one or more wired or wireless network interfaces 304, one or more I / O interfaces (input / output interfaces) 305, and one or more external devices 306 (e.g., keyboards) for communication. It may also communicate with one or more devices that enable user interaction with the device, and / or with any device that enables the device to communicate with one or more other computing devices (e.g., routers, network switches, etc.). This communication can be performed through I / O interfaces 305. Furthermore, the device can also communicate with one or more networks (e.g., local area networks (LANs)) through wired or wireless interfaces 304.

[0056] Figure 3 Only an LCD display brightness control device with components is shown; those skilled in the art will understand that... Figure 3 The structure shown does not constitute a limitation on the LCD display brightness control device, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0057] Specifically, in this embodiment, the LCD display brightness control device includes a memory and one or more programs, wherein one or more programs are stored in the memory, and one or more programs may include one or more modules, and each module may include a series of computer-executable instructions for the LCD display brightness control device, and is configured to be executed by one or more processors. The one or more programs include computer-executable instructions for performing the following: A backlight module diffusion simulation module is constructed. The backlight module diffusion simulation module performs diffusion operation on the backlight brightness distribution based on the point diffusion function to output the backlight output image. The HDR reconstruction model and the local dimming model are trained using training samples. When training the HDR reconstruction model, a feature mask is generated based on the saturation discrimination condition and the feature mask is propagated during the convolution feature extraction process to suppress saturated region features. When training the local dimming model, a supervision signal is generated by the backlight module diffusion simulation module in the absence of physical backlight output samples, and the model parameters are updated using a loss function that includes image fidelity terms and power consumption constraints. During the display stage, an input low dynamic range image is acquired, the brightness of the input low dynamic range image is corrected, and the brightness-corrected image and the feature mask are input into the trained HDR reconstruction model to output the target HDR image. The target HDR image is input into the trained local dimming model to output a backlight prediction image, and the backlight prediction image is input into the backlight module diffusion simulation module to obtain the backlight output image; The backlight output image is partitioned and converged according to the partition topology of the direct-lit backlight module to obtain a backlight control image, and a liquid crystal panel control image is generated according to the target HDR image and the backlight output image. The liquid crystal panel control image represents the pixel transmittance of the liquid crystal panel. The backlight control image drives the luminous intensity of each zone of the direct-lit backlight module, and the liquid crystal panel control image drives the transmittance of each pixel of the liquid crystal panel, so that the display device outputs a display image corresponding to the target HDR image in a backlight and liquid crystal co-modulation mode.

[0058] Based on the same idea, exemplary embodiments of the present invention also provide a computer-readable storage medium having a program product stored thereon capable of implementing the methods described above in this specification. In some possible embodiments, various aspects of this disclosure can also be implemented as a program product including program code, which, when the program product is run on a terminal device, causes the terminal device to perform the steps according to the various exemplary embodiments of this disclosure described in the "Exemplary Methods" section above.

[0059] refer to Figure 4As shown, a program 400 for implementing the above-described method according to an exemplary embodiment of the present disclosure is described. This program may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, system, or device.

[0060] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections with one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0061] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, system, or device.

[0062] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, CSS, and HTML, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0063] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal system, or network device, etc.) to execute the method according to the exemplary embodiments of this disclosure.

[0064] Furthermore, the above figures are merely illustrative representations of the processes included in the methods according to exemplary embodiments of this disclosure, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0065] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to exemplary embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0066] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0067] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for controlling the brightness of an LCD display, characterized in that, The method is applied to a display device including a direct-lit backlight module and a liquid crystal panel, wherein the direct-lit backlight module has a multi-zone light source with independently adjustable brightness, and the method includes: A backlight module diffusion simulation module is constructed. The backlight module diffusion simulation module performs diffusion operation on the backlight brightness distribution based on the point diffusion function to output the backlight output image. The HDR reconstruction model and the local dimming model are trained using training samples. When training the HDR reconstruction model, a feature mask is generated based on the saturation discrimination condition and the feature mask is propagated during the convolution feature extraction process to suppress saturated region features. When training the local dimming model, a supervision signal is generated by the backlight module diffusion simulation module in the absence of physical backlight output samples, and the model parameters are updated using a loss function that includes image fidelity terms and power consumption constraints. During the display stage, an input low dynamic range image is acquired, the brightness of the input low dynamic range image is corrected, and the brightness-corrected image and the feature mask are input into the trained HDR reconstruction model to output the target HDR image. The target HDR image is input into the trained local dimming model to output a backlight prediction image, and the backlight prediction image is input into the backlight module diffusion simulation module to obtain the backlight output image; The backlight output image is partitioned and converged according to the partition topology of the direct-lit backlight module to obtain a backlight control image, and a liquid crystal panel control image is generated according to the target HDR image and the backlight output image. The liquid crystal panel control image represents the pixel transmittance of the liquid crystal panel. The backlight control image drives the luminous intensity of each zone of the direct-lit backlight module, and the liquid crystal panel control image drives the transmittance of each pixel of the liquid crystal panel, so that the display device outputs a display image corresponding to the target HDR image in a backlight and liquid crystal co-modulation mode.

2. The LCD display brightness control method according to claim 1, characterized in that, The feature mask generation based on saturation discrimination conditions includes: The pixel brightness of the input low dynamic range image is normalized and compared with a preset saturation threshold; when the pixel brightness is lower than the preset saturation threshold, a larger mask weight is set for the corresponding pixel. When the pixel brightness reaches or exceeds the preset saturation threshold, the mask weight is monotonically compressed according to the degree to which the pixel brightness approaches the maximum brightness, so that the pixel with higher saturation corresponds to a smaller mask weight, thereby suppressing its influence on feature extraction without completely discarding the information of the saturated region. The brightness correction includes performing nonlinear gamma correction on the input low dynamic range image to adjust the brightness distribution; The output target HDR image includes: using the feature mask as weight, performing pixel-by-pixel weighted fusion of the brightness-corrected image and the predicted HDR components output by the HDR reconstruction model, so that the fusion result of the normally exposed area mainly comes from the brightness-corrected image, and the fusion result of the saturated area mainly comes from the predicted HDR components.

3. The LCD display brightness control method according to claim 1, characterized in that, The HDR reconstruction model performs feature masking operations at each convolutional layer, including: multiplying the feature map of the convolutional layer element-wise with the feature mask of the corresponding layer to obtain a mask feature map, and then inputting the mask feature map into the convolutional layer to generate the feature map of the next layer; and constructing a normalized weight kernel based on the absolute value of the weights of the convolutional kernel of the convolutional layer, and performing convolution propagation of the normalized weight kernel with the feature mask of the corresponding layer to obtain the feature mask of the next layer, wherein a preset stabilizing term is introduced in the normalization process.

4. The LCD display brightness control method according to claim 1, characterized in that, The HDR reconstruction model includes a U-shaped structure composed of an encoder and a decoder. The encoder extracts multi-scale features by downsampling step by step, and the decoder restores spatial resolution by upsampling step by step. Skip connections are set between each corresponding scale to fuse encoded features and decoded features. The encoder's convolutional layers employ a leaky rectified activation function or an equivalent nonlinear activation function, the decoder's convolutional layers employ a rectified activation function, and the output layer employs linear activation to output continuous HDR values; the upsampling employs neighbor interpolation, bilinear interpolation, or a combination thereof, and the feature mask operation is integrated in at least a portion of the convolutional layers of the encoder and the decoder.

5. The LCD display brightness control method according to claim 1, characterized in that, The loss function used to train the HDR reconstruction model includes: HDR reconstruction loss is used to calculate the aggregate amount of the absolute value of the pixel difference between the saturated region of the target HDR image and the corresponding region of the reference HDR image in the logarithmic domain, wherein the saturated region is selected to participate in the calculation through the complementary weights of the feature mask; The perceptual loss includes feature loss and style loss. The feature loss is obtained by compressing the target HDR image and the reference HDR image into a pre-trained feature extraction network to obtain multi-layer feature maps, and then aggregating the differences between the corresponding layer feature maps. The style loss is obtained by calculating Gram matrices reflecting channel correlation for each of the multi-layer feature maps, and then aggregating the differences between the corresponding layer Gram matrices. Specifically, the HDR reconstruction loss and the perception loss are combined with preset weights to update the HDR reconstruction model parameters.

6. The LCD display brightness control method according to claim 1, characterized in that, The local dimming model is a segmentation network with an encoder-decoder structure or an improved network thereof. The encoder includes a multi-level convolutional module composed of residual connections to enhance gradient propagation. The decoder includes a step-by-step upsampling module and establishes cross-layer connections with the corresponding level features of the encoder through concatenation or addition. The input of the local dimming model is obtained by combining the color channels of the target HDR image, the brightness channel calculated from the target HDR image, and the ideal backlight image. The output of the local dimming model is a backlight prediction image with the same spatial resolution as the target HDR image, and the backlight prediction value is limited to the driveable brightness range by a bounded mapping function at the output end.

7. The LCD display brightness control method according to claim 1, characterized in that, The backlight module diffusion simulation module includes: A preset point diffusion function kernel is used to characterize the light diffusion properties of the diffuser plate; The backlight prediction image is convolved with the point spread function kernel to obtain the backlight output image, and mirror filling, copy filling or periodic filling are used at the convolution boundary to reduce boundary artifacts. The partition aggregation includes: dividing the backlight output image into multiple regions corresponding to the partitions based on the number and location of the partitions of the direct-lit backlight module, and performing averaging, weighted averaging or statistical robust aggregation on the pixel brightness of each region to generate control values ​​for the corresponding partitions, thereby forming the backlight control image.

8. The LCD display brightness control method according to claim 1, characterized in that, The generation of the liquid crystal panel control image includes: Using the target HDR image as the desired display brightness distribution and the backlight output image as the incident backlight brightness distribution, the transmittance is obtained by solving pixel by pixel, so that the result of multiplying the transmittance by the backlight output image pixel by pixel approximates the target HDR image. The obtained transmittance is clipped or bounded mapping to meet the transmittance range constraints that the liquid crystal pixel can achieve; in color processing, the transmittance is solved separately on multiple color channels or uniformly on the luminance channel and combined with color ratio recovery to reduce color distortion and maintain consistency between highlight and shadow details.

9. The LCD display brightness control method according to claim 1, characterized in that, The image fidelity items include: The difference between the simulated HDR image and the target HDR image in the pixel domain or the perceptual domain is measured to obtain a regression loss. The regression loss is used to drive the backlight prediction image output by the local dimming model and the liquid crystal panel control image to reproduce the target HDR image under co-modulation. The power consumption constraint includes: The overall brightness level of the backlight prediction image is aggregated and added as a penalty term to the loss function to suppress excessively bright backlight output and reduce backlight energy consumption.

10. An LCD display brightness control system, characterized in that, include: A module is used to construct a backlight module diffusion simulation module. The backlight module diffusion simulation module performs diffusion operations on the backlight brightness distribution based on the point diffusion function to output a backlight output image. The model training module is used to train the HDR reconstruction model and the local dimming model using training samples. When training the HDR reconstruction model, a feature mask is generated based on the saturation discrimination condition and the feature mask is propagated during the convolution feature extraction process to suppress saturated region features. When training the local dimming model, a supervision signal is generated by the backlight module diffusion simulation module in the absence of physical backlight output samples, and the model parameters are updated using a loss function that includes image fidelity terms and power consumption constraints. The correction module is used to acquire the input low dynamic range image during the display stage, perform brightness correction on the input low dynamic range image, and input the brightness-corrected image and the feature mask into the trained HDR reconstruction model to output the target HDR image. The prediction module is used to input the target HDR image into the trained local dimming model to output a backlight prediction image, and input the backlight prediction image into the backlight module diffusion simulation module to obtain the backlight output image; A brightness control module is used to partition and converge the backlight output image according to the partition topology of the direct-lit backlight module to obtain a backlight control image, and generate a liquid crystal panel control image according to the target HDR image and the backlight output image, wherein the liquid crystal panel control image represents the pixel transmittance of the liquid crystal panel. The display control module is used to drive the luminous intensity of each zone of the direct-lit backlight module according to the backlight control image, and to drive the transmittance of each pixel of the liquid crystal panel according to the liquid crystal panel control image, so that the display device outputs a display image corresponding to the target HDR image in a backlight and liquid crystal co-modulation mode.

Images