Backlight fusion convolutional neural network image quality enhancement method and system, terminal and medium

Abstract

The invention provides a backlight fusion convolutional neural network image quality enhancement method and system, a terminal and a medium, and relates to the technical field of display, and the method comprises the steps: extracting backlight data of a logic partition number corresponding to a to-be-displayed image; inputting the data into a pre-trained target backlight fusion model, and obtaining backlight data of which the backlight partition number is smaller than the actual partition number of the logic partition number after the model fuses the backlight data of the logic partition number; transmitting the backlight data of the actual partition number and the to-be-displayed image to preset display equipment for display; the display effect of the backlight data of the actual partition number approaches the display effect of the backlight data of the logic partition number. The high-partition-number backlight data is fused into the low-partition-number backlight data through the model, the image quality enhancement effect close to that of a high-backlight-partition-number backlight module is achieved on the low-backlight-partition-number backlight module, the backlight partition number is reduced under the condition that the display performance is not lost, and the manufacturing cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of display technology, and in particular to a backlight fusion convolutional neural network image quality enhancement method, system, terminal and medium. Background Art

[0002] Mini LED technology is widely used in the current high-end, large-screen TV market, driven by the pursuit of high-contrast, high-quality displays. However, in the current local dimming field, achieving greater image quality often requires a larger number of backlight zones, leading to increased manufacturing costs. Furthermore, as the number of different backlight zone options increases, the local dimming algorithm and LED backlight module require more time and accurate debugging to achieve optimal results.

[0003] In summary, related technologies often achieve better image quality enhancement by increasing the number of backlight zones, which results in high manufacturing costs and complex and time-consuming debugging. Therefore, how to provide a solution to the above technical problems is a problem currently needed by those skilled in the art. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that, in response to the above-mentioned defects of the prior art, a backlight fusion convolutional neural network image quality enhancement method, system, terminal and medium are provided, aiming to solve the problems of high manufacturing cost and complex and time-consuming debugging when obtaining better image quality enhancement effect by increasing the number of backlight partitions in the prior art.

[0005] The technical solutions adopted by the present invention to solve the technical problems are as follows: In a first aspect, a backlight fusion convolutional neural network image quality enhancement method is provided, wherein the method comprises: Acquire an image to be displayed, and extract backlight data of a logical partition number corresponding to the image to be displayed; Inputting the backlight data of the logical partition number into a pre-trained target backlight fusion model to obtain the backlight data of the actual partition number output by the target backlight fusion model after fusing the backlight data of the logical partition number; Transmitting the backlight data of the actual number of partitions and the image to be displayed to a preset display device for display; The actual number of partitions is smaller than the logical number of partitions, and the similarity between the display effect of the backlight data of the actual number of partitions output by the target backlight fusion model and the display effect of the backlight data of the logical number of partitions meets a preset similarity condition.

[0006] Optionally, the training process of the pre-trained target backlight fusion model includes: A set of image sample data is selected from a pre-built training set as training samples, and a current image sample and a backlight sample of the corresponding logical partition number are obtained; Inputting the backlight samples of the logical partition number corresponding to the current image sample into a backlight fusion model constructed based on a convolutional neural network, and obtaining the backlight samples of the actual partition number output by the backlight fusion model; Performing display effect simulation on the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, respectively, to obtain a first simulation result corresponding to the backlight samples of the logical partition number and a second simulation result corresponding to the backlight samples of the actual partition number; Fine-tuning the parameters of the backlight fusion model according to the first simulation result and the second simulation result to obtain a current fine-tuned backlight fusion model; Determining whether all image sample data in the training set are selected as training samples; When all image sample data in the training set are selected as training samples, it is determined that one training of the backlight fusion model has been completed, and the current training number of the backlight fusion model is determined, and it is determined whether the current training number reaches a preset training number; When the current training times reaches the preset training times, the currently fine-tuned backlight fusion model is determined as the trained target backlight fusion model.

[0007] Optionally, after determining whether all image sample data in the training set are selected as training samples, the method further includes: When all the image sample data in the training set have not been selected as training samples, it is determined that one training of the backlight fusion model has not been completed, and the steps of selecting a group of image sample data from the pre-constructed training set as training samples, obtaining the current image sample and the backlight samples of the corresponding logical partition number, and subsequent steps are continued until all the image sample data in the training set have been selected as training samples; Furthermore, after determining whether the current number of training times reaches the preset number of training times, the method further includes: When the current number of training times does not reach the preset number of training times, the step of selecting a set of image sample data from the pre-constructed training set as training samples to obtain the current image sample and the backlight sample of its corresponding logical partition number and its subsequent steps are re-executed until the current number of training times reaches the preset number of training times.

[0008] Optionally, the performing display effect simulation on the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, respectively, to obtain a first simulation result corresponding to the backlight samples of the logical partition number and a second simulation result corresponding to the backlight samples of the actual partition number, includes: Performing diffusion processing on the backlight samples of the logical partition number to obtain a first diffusion result; Performing pixel compensation on the image sample according to the first diffusion result to obtain a first compensated image sample; Performing a display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters, and the backlight samples of the logical partition number, to obtain a first simulation result corresponding to the backlight samples of the logical partition number; Performing a display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters, and the backlight samples of the actual number of partitions, to obtain a second simulation result corresponding to the backlight samples of the actual number of partitions; Alternatively, diffusion processing is performed on the backlight samples of the logical partition number to obtain a first diffusion result, and diffusion processing is performed on the backlight samples of the actual partition number to obtain a second diffusion result; Performing pixel compensation on the image sample according to the first diffusion result to obtain a first compensated image sample, and performing pixel compensation on the image sample according to the second diffusion result to obtain a second compensated image sample; Performing a display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters, and the backlight samples of the logical partition number, to obtain a first simulation result corresponding to the backlight samples of the logical partition number; The display effect of the current image sample is simulated according to the pixel values ​​corresponding to each pixel point on the second compensated image sample, the liquid crystal module parameters and the backlight samples of the actual partition number, and the second simulation result corresponding to the backlight samples of the actual partition number is obtained.

[0009] Optionally, performing diffusion processing on the backlight samples of the logical partition number to obtain a first diffusion result includes: Performing diffusion processing on the backlight samples of the logical partition number by using anisotropic Gaussian blur diffusion method to obtain a first diffusion result; Furthermore, performing diffusion processing on the backlight samples of the actual number of partitions to obtain a second diffusion result includes: The backlight samples of the actual number of partitions are diffused using an isotropic Gaussian blur diffusion method to obtain a second diffusion result.

[0010] Optionally, the convolutional neural network includes an average pooling layer, four convolutional layers and an activation function layer corresponding to each convolutional layer, and the first convolutional layer and its corresponding activation function layer serve as the input layer, the second convolutional layer and its corresponding activation function layer, the third convolutional layer and its corresponding activation function layer and the average pooling layer serve as hidden layers, and the fourth convolutional layer and its corresponding activation function layer serve as the output layer.

[0011] Optionally, the calculation formula of the convolutional layer is: ; ; in, is the output channel index, Represents the convolutional layer The convolution result of the output channel at the (i, j) coordinate, i is the row index of the convolution layer input data or output data, j is the column index of the convolution layer input data or output data, Indicates the The offset value of each output channel, is the input channel index, is the convolution kernel size, u is the row index of the convolution kernel, v is the column index of the convolution kernel, is the input data of the c-th input channel of the convolutional layer at the coordinate (i+u, j+v), Represents the weight data of the c-th channel of the convolution kernel at the (u, v) coordinate; Furthermore, the calculation formula of the average pooling layer is: ; in, is the pooling result of the average pooling layer, The row index of the average pooling layer output data, is the column index of the average pooling layer output data, H is the height of the pooling window, W is the width of the pooling window, is the input data of the average pooling layer at the coordinate (x+h, y+w), x is the row index of the average pooling layer input data, and y is the column index of the average pooling layer input data.

[0012] In a second aspect, the present invention further discloses a backlight fusion convolutional neural network image quality enhancement system, wherein the system comprises: An acquisition module, configured to acquire an image to be displayed and extract backlight data of a logical partition number corresponding to the image to be displayed; A fusion module, configured to input the backlight data of the logical partition number into a pre-trained target backlight fusion model, and obtain the backlight data of the actual partition number output by the target backlight fusion model after fusing the backlight data of the logical partition number; A display module, configured to transmit the backlight data of the actual number of partitions and the image to be displayed to a preset display device for display; The actual number of partitions is smaller than the logical number of partitions, and the similarity between the display effect of the backlight data of the actual number of partitions output by the target backlight fusion model and the display effect of the backlight data of the logical number of partitions meets a preset similarity condition.

[0013] In a third aspect, the present invention discloses a terminal, which includes: a memory, a processor, and a backlight fusion convolutional neural network image quality enhancement program stored in the memory and runnable on the processor, wherein the backlight fusion convolutional neural network image quality enhancement program implements the steps of the backlight fusion convolutional neural network image quality enhancement method as described above when executed by the processor.

[0014] In a fourth aspect, the present invention discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program can be executed to implement the steps of the backlight fusion convolutional neural network image quality enhancement method as described above.

[0015] The present invention provides a backlight fusion convolutional neural network image quality enhancement method, system, terminal and medium, and the backlight fusion convolutional neural network image quality enhancement method includes: obtaining an image to be displayed, and extracting backlight data of a logical partition number corresponding to the image to be displayed; inputting the backlight data of the logical partition number into a pre-trained target backlight fusion model to obtain the backlight data of the actual partition number output after the target backlight fusion model fuses the backlight data of the logical partition number; transmitting the backlight data of the actual partition number and the image to be displayed to a preset display device for display; wherein the actual partition number is less than the logical partition number, and the similarity between the display effect of the backlight data of the actual partition number output by the target backlight fusion model and the display effect of the backlight data of the logical partition number meets a preset similarity condition. It can be seen from this that the present invention can fuse the backlight data of the logical partition number into the backlight data of the actual partition number through the pre-trained target backlight fusion model, and the actual partition number is smaller than the logical partition number, so that the image quality enhancement effect close to that of the backlight module with a high backlight partition number can be achieved on the backlight module with a low backlight partition number, that is, the number of backlight partitions can be reduced without losing display performance, thereby reducing the manufacturing cost to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of a preferred embodiment of the backlight fusion convolutional neural network image quality enhancement method of the present invention; Figure 2 This is a schematic diagram of a specific backlight fusion model training framework disclosed in the present invention; Figure 3 This is a schematic diagram of a specific display system structure disclosed in the present invention; Figure 4 This is a specific backlight fusion model training flow chart disclosed in the present invention; Figure 5 This is a schematic diagram of the luminous effect of a partitioned backlight LED disclosed in the present invention; Figure 6 It is a structural diagram of a specific convolutional neural network disclosed in the present invention; Figure 7 This is another specific backlight fusion model training flow chart disclosed in the present invention; Figure 8 This is a schematic diagram of a specific correspondence between high-partition number backlight data and low-partition number backlight data disclosed in the present invention; Figure 9 This is another schematic diagram of the backlight LED lighting effect of the partition number disclosed in the present invention; Figure 10 This is a functional principle block diagram of a preferred embodiment of the backlight fusion convolutional neural network image quality enhancement system of the present invention; Figure 11 It is a functional principle block diagram of a preferred embodiment of the terminal in the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clear and distinct, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] In the current high-end, large-screen TV market, Mini LED technology is widely used in pursuit of high-contrast, high-quality displays. However, in the current local dimming field, achieving better image quality often requires a larger number of backlight zones, leading to increased manufacturing costs. Furthermore, as the number of different backlight zone options increases, the local dimming algorithm and the LED backlight module require time-consuming and accurate debugging to achieve optimal results. This means that related technologies often achieve better image quality by increasing the number of backlight zones, resulting in high manufacturing costs and complex, time-consuming debugging.

[0019] However, with the development of artificial intelligence technology, especially convolutional neural networks (CNN), it has demonstrated excellent performance in various industries, especially in image processing-related industries. To this end, this application provides a backlight fusion convolutional neural network image quality enhancement solution that can reduce the number of backlight partitions without sacrificing display performance, which can reduce manufacturing costs to a certain extent.

[0020] See Figure 1 , Figure 1 This is a flow chart of the backlight fusion convolutional neural network image quality enhancement method in the present invention. Figure 1 As shown, the backlight fusion convolutional neural network image quality enhancement method described in the embodiment of the present invention includes: Step S11: Acquire an image to be displayed, and extract backlight data of a logical partition number corresponding to the image to be displayed.

[0021] It can be understood that for the image to be displayed, the backlight data of the corresponding logical partition number, that is, the corresponding high partition number backlight data, is extracted, and specifically, the backlight data of the logical partition number corresponding to the image to be displayed (high partition number backlight data) can be extracted using an artificial intelligence-based backlight extraction algorithm.

[0022] Step S12: input the backlight data of the logical partition number into a pre-trained target backlight fusion model to obtain the backlight data of the actual partition number output by the target backlight fusion model after fusing the backlight data of the logical partition number; wherein, the actual number of partitions is smaller than the logical number of partitions, and the similarity between the display effect of the backlight data of the actual number of partitions output by the target backlight fusion model and the display effect of the backlight data of the logical partition number meets a preset similarity condition.

[0023] In this embodiment, after extracting the backlight data of the logical partition number corresponding to the image to be displayed, the backlight data of the logical partition number can be input into a pre-trained target backlight fusion model, so that the target backlight fusion model can be used to fuse the backlight data of the logical partition number into the backlight data of the actual partition number (low partition number backlight data). It can be understood that the pre-trained target backlight fusion model can fuse the backlight data of the logical partition number corresponding to the image to be displayed into backlight data of an actual partition number, wherein the actual partition number is less than the logical partition number, and the similarity between the display effect of the backlight data of the actual partition number output by the target backlight fusion model and the display effect of the backlight data of the logical partition number meets the preset similarity condition, such as the similarity between the display effects corresponding to the two is not less than 95%, that is, the pre-trained target backlight fusion model can fuse high-partition-number backlight data into relatively low-partition-number backlight data, and can achieve image quality enhancement effects close to those of backlight modules with high backlight partition numbers on backlight modules with low backlight partition numbers, so that high-partition backlight extraction algorithms with better performance can quickly adapt to backlight modules with different low backlight partition numbers, thereby accelerating product development speed, that is, it can effectively fuse backlight data, so that different logical backlight extraction algorithms (obtaining high-partition-number backlight data) and different actual backlight schemes (using low-partition-number backlight data) can adapt to each other.

[0024] Among them, during the training process of the target backlight fusion model, the difference between the display effect simulation results of the backlight data of the logical partition number and the display effect simulation results of the backlight data of the actual partition number is compared, and then the model parameters are adjusted until after a preset number of training times, the display effect simulation results of the backlight data of the actual partition number are close to the display effect simulation results of the backlight data of the logical partition number, and then the training is stopped, that is, the display effect simulation results of the backlight data with a high partition number and the display effect simulation results of the backlight data with a low partition number are compared and learned, so that the display effect of the backlight data with a low partition number is close to the display effect of the backlight data with a high partition number.

[0025] For example, see Figure 2As shown, the high-partition backlight data corresponding to the RGB image sample is extracted through the high-partition backlight extraction algorithm, and the high-partition backlight data is input into the backlight fusion model to fuse the high-partition backlight data into low-partition backlight data through the backlight fusion model. Then, the high-partition simulator is used to simulate the display effect of the high-partition backlight data to obtain the corresponding high-partition backlight simulation image. The low-partition simulator is used to simulate the display effect of the low-partition backlight data to obtain the corresponding low-partition backlight simulation image. The difference between the high-partition backlight simulation image and the low-partition backlight simulation image is compared to adjust the parameters of the backlight fusion model, such as weight data and bias data. After adjusting the model parameters, the above process is repeated until the preset number of training times is met and the display effect of the low-partition backlight simulation image approaches the display effect of the high-partition backlight simulation image, thereby The target parameters of the backlight fusion model are determined, and the target parameters are used as the final parameters of the backlight fusion model to obtain the target backlight fusion model. That is, in the process of pre-training the backlight fusion model constructed based on the convolutional neural network, by comparing the difference between the display simulation results of the low-partition-number backlight data and the display simulation results of the high-partition-number backlight data, the difference between the display simulation results before and after the fusion of the high-partition-number backlight data is learned, so that the display simulation results of the low-partition-number backlight data gradually approach the display simulation results of the high-partition-number backlight data, and finally the display effect of the low-partition-number backlight data approaches the display effect of the high-partition-number backlight data, thereby enhancing the image quality and ensuring the display effect without increasing the number of backlight partitions. That is, without losing the display performance, the number of actual backlight module partitions is reduced, thereby reducing the manufacturing cost.

[0026] Step S13: transmitting the backlight data of the actual number of partitions and the image to be displayed to a preset display device for display.

[0027] In this embodiment, after the target backlight fusion model fuses the backlight data of the logical number of partitions into the backlight data of the actual number of partitions, the backlight data of the actual number of partitions and the image to be displayed can be directly transmitted to the preset display device for display. Moreover, the preset display device can be a Mini LED display, for example, the backlight data of the actual number of partitions and the image to be displayed can be transmitted to the Mini LED display for display.

[0028] In this embodiment, before transmitting the backlight data of the actual number of partitions and the image to be displayed to the preset display device for display, it can also specifically include: correcting the image to be displayed to obtain a corrected image to be displayed, and then transmitting the backlight data of the actual number of partitions and the corrected image to be displayed to the preset display device for display.

[0029] It should be noted that the above-mentioned backlight fusion convolutional neural network image quality enhancement method can be implemented using platforms such as CPU (Central Processing Unit), GPU (Graphics Processing Unit, also known as graphics card), NPU (Neural Processing Unit, neural network processor) or FPGA (Field-Programmable Gate Array, field programmable gate array), for example, see Figure 3 As shown, a pre-trained target backlight fusion model is deployed on the GPU as a submodule and used in conjunction with the CPU and Mini LED display to form a complete display system. This system can obtain input image data via methods such as HDMI (High-Definition Multimedia Interface) or DP (DisplayPort, a digital video interface standard). The CPU extracts the high-partition backlight data corresponding to the input image data and performs certain corrections on the input image data in the central processing unit to obtain output image data. The high-partition backlight data is then transmitted to the GPU on which the target backlight fusion model is deployed for processing, obtaining low-partition backlight data. The low-partition backlight data and the output image data are then transmitted to the Mini LED display, which includes an image display driver circuit and an LED backlight driver circuit, for display. For another example, a pre-trained target backlight fusion model is deployed on an FPGA system. The FPGA system receives high-partition backlight data from a SoC (System on Chip), fuses the high-partition backlight data, and obtains low-partition backlight data. The low-partition backlight data is then output and used together with the SoC output image data to control the TV display. Among them, a resource-constrained FPGA chip is selected to complete the deployment. Therefore, for the convolution layer, only two convolution kernels are implemented. Different layers of convolution can be calculated using time-division multiplexing. Since the backlight data is relatively small, it is fully possible to complete the calculation and output within the time of one frame of image data.

[0030] For the specific training process of the pre-trained target backlight fusion model, see Figure 4 As shown, the following steps may be specifically included: Step S121 : Select a group of image sample data from a pre-built training set as training samples, and obtain the current image sample and the backlight samples of the corresponding logical partition number.

[0031] In this embodiment, the pre-constructed training set contains multiple groups of image sample data, and each group of image sample data contains image samples and backlight samples of their corresponding logical partition numbers. Then, during each training, a group of image sample data is randomly selected from the training set as the current training sample data.

[0032] Step S122: input the backlight samples of the logical partition number corresponding to the current image sample into the backlight fusion model constructed based on the convolutional neural network to obtain the backlight samples of the actual partition number output by the backlight fusion model.

[0033] In this embodiment, a backlight fusion model constructed based on a convolutional neural network is used to fuse the backlight samples of the logical partition number (high partition number backlight samples) corresponding to the current image sample into the backlight samples of the actual partition number (low partition number backlight samples), and the actual partition number is less than the logical partition number.

[0034] Step S123: Simulate the display effect of the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, and obtain a first simulation result corresponding to the backlight samples of the logical partition number and a second simulation result corresponding to the backlight samples of the actual partition number.

[0035] In this embodiment, after the backlight fusion model outputs the backlight samples of the actual partition number corresponding to the backlight samples of the logical partition number, the display effect simulation is performed on the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, that is, the display effect simulation is performed on the backlight samples of the logical partition number and the backlight samples of the actual partition number using the current image sample, respectively, to obtain a first simulation result corresponding to the backlight samples of the logical partition number and a second simulation result corresponding to the backlight samples of the actual partition number.

[0036] In a specific embodiment, the display effect simulation is performed on the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, which may specifically include: performing diffusion processing on the backlight samples of the logical partition number to obtain a first diffusion result; performing pixel compensation on the image sample according to the first diffusion result to obtain a first compensated image sample; performing display effect simulation on the current image sample according to the pixel values ​​corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters and the backlight samples of the logical partition number to obtain a first simulation result corresponding to the backlight samples of the logical partition number; performing display effect simulation on the current image sample according to the pixel values ​​corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters and the backlight samples of the actual partition number to obtain a second simulation result corresponding to the backlight samples of the actual partition number.

[0037] In another specific embodiment, the display effect simulation is performed on the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, which may specifically include: performing diffusion processing on the backlight samples of the logical partition number to obtain a first diffusion result, and performing diffusion processing on the backlight samples of the actual partition number to obtain a second diffusion result; performing pixel compensation on the image sample according to the first diffusion result to obtain a first compensated image sample, and performing pixel compensation on the image sample according to the second diffusion result to obtain a second compensated image sample; performing display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters and the backlight samples of the logical partition number, to obtain a first simulation result corresponding to the backlight sample of the logical partition number; performing display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the second compensated image sample, the liquid crystal module parameters and the backlight samples of the actual partition number, to obtain a second simulation result corresponding to the backlight sample of the actual partition number.

[0038] It is understandable that during the model training process, when simulating the display effect, if the input image sample cannot be obtained, the image sample can be pixel-compensated only according to the diffusion result of the high-partition backlight data. This pixel compensation is usually completed by the upstream device that extracts the high-partition backlight sample. The high-partition backlight sample and the low-partition backlight sample are simulated based on this compensation result. When the input image sample is obtained, pixel compensation is performed according to the first diffusion result of the high-partition backlight sample and the second diffusion result of the low-partition backlight sample to obtain the corresponding compensation results. The high-partition backlight sample and the low-partition backlight sample can be simulated according to the corresponding compensation results. The backlight data is used to illuminate the backlight module, and the compensated image data is sent to the liquid crystal module. Therefore, the display effect simulation is completed by combining the backlight data, image data, compensated image data, and liquid crystal module parameters.

[0039] In addition, the backlight samples of the logical partition number are diffused to obtain a first diffusion result, which may specifically include: using anisotropic Gaussian blur diffusion method to diffuse the backlight samples of the logical partition number to obtain the first diffusion result; and the backlight samples of the actual partition number are diffused to obtain a second diffusion result, which may specifically include: using isotropic Gaussian blur diffusion method to diffuse the backlight samples of the actual partition number to obtain the second diffusion result.

[0040] It is understandable that since the lamp beads will affect each other, the backlight samples need to be diffused to simulate the real situation, and different diffusion coefficients need to be used according to the real situation. For example, for the diffusion of backlight data with a low number of partitions, the isotropic Gaussian blur diffusion method is used for diffusion, and for the backlight data with a high number of partitions, the anisotropic Gaussian blur diffusion method is used for diffusion to simulate different lamp bead densities in the horizontal and vertical directions.

[0041] It should be noted that different backlight schemes have different numbers of partitions and different backlight module parameters, such as diffusion coefficient and OD (Optical Distance) value, which will lead to different display effects. Figure 5 As shown, simulation processing can be performed according to actual conditions. In this embodiment, the proportional relationship between the backlight data with a high partition number and the backlight data with a low partition number and the backlight module parameters are used to calculate the diffusion coefficient of a single backlight partition in the horizontal and vertical directions, and the Gaussian blur kernel is used to simulate the diffusion of the backlight data. The display effect is simulated by combining the liquid crystal module parameters, such as transmittance, and the visual effect of the human eye.

[0042] It should also be pointed out that since most backlight modules cannot drive all lamp beads at full power, it is also possible to limit the power of the backlight sample before simulating the display effect to ensure that the module does not operate at overpower. The backlight data will be converted into the current that drives the backlight partitions (lamp beads). The power provided by the power supply is limited, so it is necessary to ensure that the total power of all backlight partitions does not exceed the power provided by the power supply.

[0043] Step S124 : fine-tune the parameters of the backlight fusion model according to the first simulation result and the second simulation result to obtain a currently fine-tuned backlight fusion model.

[0044] It is understandable that by comparing the difference between the first simulation result and the second simulation result, the parameters of the backlight fusion model are adjusted so that the second simulation result of the low partition number backlight sample gradually approaches the first simulation result of the high partition number backlight sample.

[0045] Step S125: Determine whether all image sample data in the training set are selected as training samples.

[0046] It is understandable that due to the possibility that the same training sample may be selected multiple times in random selection, a training sample can be selected from the training set once, and after using the training sample to complete the adjustment of the backlight fusion model, it is confirmed again whether each set of image sample data in the training set has been used for training, to ensure that in the current round of training, all training samples in the training set are selected to train the backlight fusion model.

[0047] Step S126: When all image sample data in the training set are selected as training samples, it is determined that one training of the backlight fusion model has been completed, and the current training times of the backlight fusion model are determined, and it is determined whether the current training times reach the preset training times.

[0048] It can be understood that in each training round, all the image sample data in the training set will be used to adjust the backlight fusion model, and when all the image sample data in the training set are selected as training samples, the backlight fusion model will be trained once. At this time, the number of training times is increased by one to determine the current number of training times and to judge whether the current number of training times reaches the preset number of training times.

[0049] Step S127: When the current training times reaches the preset training times, the current fine-tuned backlight fusion model is determined as the trained target backlight fusion model.

[0050] It can be understood that after a preset number of training times, when the display effect simulation results of the backlight samples with the actual number of partitions are close to the display effect simulation results of the backlight samples with the logical number of partitions, the model training is stopped to obtain the target backlight fusion model.

[0051] In this embodiment, after determining whether all the image sample data in the training set have been selected as training samples, it may further specifically include: when all the image sample data in the training set have not been selected as training samples, it is determined that one training of the backlight fusion model has not been completed, and the steps of selecting a group of image sample data from the pre-constructed training set as training samples, obtaining the current image sample and the backlight sample corresponding to the number of logical partitions, and subsequent steps thereof, are continued until all the image sample data in the training set are selected as training samples; and, after determining whether the current number of training times has reached the preset number of training times, it may further specifically include: when the current number of training times has not reached the preset number of training times, the steps of selecting a group of image sample data from the pre-constructed training set as training samples, obtaining the current image sample and the backlight sample corresponding to the number of logical partitions, and subsequent steps thereof, are re-executed until the current number of training times reaches the preset number of training times.

[0052] It should be noted that, see Figure 6 As shown in the figure, the convolutional neural network includes an average pooling layer, four convolutional layers and an activation function layer corresponding to each convolutional layer, and the first convolutional layer and its corresponding activation function layer are used as the input layer, the second convolutional layer and its corresponding activation function layer, the third convolutional layer and its corresponding activation function layer and the average pooling layer are used as the hidden layer, and the fourth convolutional layer and its corresponding activation function layer are used as the output layer.

[0053] Among them, the calculation formula of the convolution layer is: ; ; in, is the output channel index, Represents the convolutional layer The convolution result of the output channel at the (i, j) coordinate, i is the row index of the convolution layer input data or output data, j is the column index of the convolution layer input data or output data, Indicates the The offset value of each output channel, is the input channel index, is the convolution kernel size, u is the row index of the convolution kernel, v is the column index of the convolution kernel, is the input data of the c-th input channel of the convolutional layer at the coordinate (i+u, j+v), Represents the weight data of the c-th channel of the convolution kernel at the (u, v) coordinate.

[0054] It should also be noted that to ensure the consistency of the input and output sizes of the convolutional layer, the input data needs to be padded. Padding methods can include, but are not limited to, zero-value padding, fixed-value padding, reflection padding, cyclic padding, and replication padding, with a total of ks circles padded. The main parameters of convolution are the convolution kernel size, number of input channels, number of output channels, padding method, and step size. Therefore, different parameter configurations can be selected to design the convolutional neural network according to actual conditions. The activation function layer can use nonlinear functions such as sigmoid, tanh, ReLU, and LeakyReLU as the activation function of the convolutional layer.

[0055] And, the calculation formula of the average pooling layer is: ; in, is the pooling result of the average pooling layer, The row index of the average pooling layer output data, is the column index of the average pooling layer output data, H is the height of the pooling window, W is the width of the pooling window, is the input data of the average pooling layer at the coordinate (x+h, y+w), x is the row index of the average pooling layer input data, and y is the column index of the average pooling layer input data.

[0056] It should be noted that the stride of the general pooling layer is consistent with the width and height of the pooling window, and the ratio of the width and height of the input data to the width and height of the output data is W and H. The design of the pooling window is consistent with the ratio of the logical backlight data (high-partition backlight data) to the actual backlight module backlight data (low-partition backlight data). The aspect ratio of the high-partition backlight data to the low-partition backlight data is W and H.

[0057] For example, see Figure 7 As shown, the overall training process of the backlight fusion model constructed based on the convolutional neural network can be specifically as follows: reading the training sample data from the pre-constructed training set, that is, reading the current image sample and its corresponding high-partition number backlight data. If there is no high-partition number backlight data, the backlight extraction algorithm can be used to extract the high-partition number backlight data corresponding to the current image sample, and if there is a need to accelerate the training process, multiple groups of sample data can be read at one time, such as batch_size groups of sample data, and training can be performed based on multiple groups of sample data at the same time. The backlight fusion model is used to fuse the high-partition number backlight data into low-partition number backlight data, and the display effect of the image is simulated based on the high-partition number backlight data and the low-partition number backlight data to obtain a first simulation result corresponding to the high-partition number backlight data and a second simulation result corresponding to the low-partition number backlight data. Fine-tune the parameters of the backlight fusion model based on the first simulation result and the second simulation result, and determine whether all sample data in the training set has been processed. If so, determine whether the backlight fusion model is being repeatedly trained N times. If so, terminate the training and use the backlight fusion model with fine-tuned parameters as the trained target backlight fusion model. If not all sample data in the training set has been processed, read the next set of sample data from the training set and continue training the backlight fusion model until the backlight fusion model has been trained N times. N is a preset number of training times.

[0058] It can be seen that in the embodiment of the present invention, the backlight data of the logical partition number can be fused into the backlight data of the actual partition number through the pre-trained target backlight fusion model, and the actual partition number is smaller than the logical partition number, so that the image quality enhancement effect close to that of the backlight module with a high backlight partition number can be achieved on the backlight module with a low backlight partition number, that is, the number of backlight partitions can be reduced without losing too much display performance, thereby reducing the manufacturing cost to a certain extent.

[0059] For example, when the resolution of the image to be displayed is 3840×2160, the corresponding number of logical partitions (high backlight partitions) can be 64×36, and the actual number of physical partitions (low backlight partitions) can be 32×18, that is, every 2×2 logical partitions corresponds to one actual physical partition. In addition, all convolutional layers in the backlight fusion model built based on the convolutional neural network can use symmetrical padding with a stride of 1. The convolution kernel size and number of input and output channels from the first convolution layer to the second convolution layer are: The convolution kernel size of the first convolution layer is 5×5, with 1 input channel and 8 output channels; The convolution kernel size of the second convolution layer is 3×3, with 8 input channels and 32 output channels; The convolution kernel size of the third convolution layer is 3×3, with 32 input channels and 8 output channels; The convolution kernel size of the fourth convolution layer is 3×3, with 8 input channels and 1 output channel.

[0060] The pooling window size of the average pooling layer is 2×2, which is consistent with the correspondence between logical partitions and actual physical partitions.

[0061] The backlight fusion model constructed based on the convolutional neural network is trained according to the above model training process. The preset training times N can be 1000 and the batch_size is 16, until the display effect of the low-partition number backlight data approaches the display effect of the high-partition number backlight data. After the training is completed, the trained target backlight fusion model can be used to fuse the high-partition number backlight data into the low-partition number backlight data, and the backlight data of the actual partition number and the image to be displayed are transmitted to the preset display device for display.

[0062] It should be noted that in traditional algorithms, the data between corresponding partitions are generally directly fused to obtain the backlight data of the low partition number. For example, the averaging operation is performed on the backlight data of the high partition number, that is: ; Among them, LED (i, j) represents the low partition number backlight data, i represents the row index of the low partition number backlight data, j represents the column index of the low partition number backlight data; LED_1, LED_2, LED_3, LED_4 represent the corresponding high partition number backlight data.

[0063] See also Figure 8As shown in the figure, in the traditional algorithm, each low-partition backlight data is fused from 2×2 high-partition backlight data, that is, each low-partition backlight data contains the information of 2×2 high-partition backlight data. However, in this application, in the pre-trained target backlight fusion model, the 3×3 convolution layer will fuse 3×3 data, the 5×5 convolution layer will fuse 5×5 data, and the 2×2 pooling layer will fuse 2×2 data. Therefore, four convolution layers and one pooling layer will fuse a total of 11 high-partition data. Compared with the traditional algorithm, more information is fused, so there is a greater probability of obtaining better image quality enhancement effects.

[0064] For example, when the resolution of the image to be displayed is 3840×2160, the corresponding number of logical partitions (high backlight partitions) can be 64×18, and the actual number of physical partitions (low backlight partitions) can be 32×18, that is, every 2×1 logical partitions corresponds to one actual physical partition. In addition, all convolutional layers in the backlight fusion model built based on the convolutional neural network can use zero padding with a stride of 1. The convolution kernel size and the number of input and output channels from the first convolution layer to the second convolution layer are: The convolution kernel size of the first convolution layer is 3×3, with 1 input channel and 2 output channels; The convolution kernel size of the second convolution layer is 3×3, with 2 input channels and 2 output channels; The convolution kernel size of the third convolution layer is 3×3, with 2 input channels and 2 output channels; The convolution kernel size of the fourth convolution layer is 3×3, with 2 input channels and 1 output channel.

[0065] The pooling window size of the average pooling layer is 2×1, which is consistent with the correspondence between logical partitions and actual physical partitions.

[0066] Similarly, the backlight fusion model constructed based on the convolutional neural network is trained according to the above model training process. The preset number of training times N can be 1000, and the batch_size can be 32, until the display effect of the low-partition number backlight data approaches the display effect of the high-partition number backlight data. After the training is completed, the trained target backlight fusion model can be used to fuse the high-partition number backlight data into the low-partition number backlight data, and the backlight data of the actual partition number and the image to be displayed are transmitted to the preset display device for display.

[0067] From the above, it can be seen that the present application utilizes the learning ability of the backlight fusion model to calculate new low-partition-number backlight data based on a larger range of high-partition-number backlight data, and can perform logical multiplication of arbitrary integer scales on rows and columns. It is more versatile and can adapt to more backlight module solutions. In the simulation of backlight solutions with different numbers of partitions, the obtained backlight data can be simulated more realistically based on the parameters of the actual backlight module and the partition data, thereby guiding the training of the backlight fusion model constructed based on the convolutional neural network, and giving full play to the learning ability of the backlight fusion model.

[0068] It should be noted that the density of lamp beads in the horizontal and vertical directions of high-zone backlight data is different, so the backlight module will make special treatments for this, such as adding corresponding optical films, designing special brackets, etc., to obtain uniform backlight. Figure 9 As shown in the figure, although the lamp beads themselves have a circular diffusion effect (i.e., the diffusion effect is consistent in the horizontal and vertical directions), the diffusion effect of the lamp beads will be inconsistent in the horizontal and vertical directions after passing through the optical film, making the actual diffusion effect of a single lamp bead inconsistent in the horizontal and vertical directions.

[0069] In one embodiment, if Figure 10 As shown, based on the above-mentioned backlight fusion convolutional neural network image quality enhancement method, the present invention also provides a backlight fusion convolutional neural network image quality enhancement system, including: An acquisition module 11 is configured to acquire an image to be displayed and extract backlight data of a logical partition number corresponding to the image to be displayed; A fusion module 12 is configured to input the backlight data of the logical partition number into a pre-trained target backlight fusion model, and obtain the backlight data of the actual partition number output by the target backlight fusion model after fusing the backlight data of the logical partition number; A display module 13 is used to transmit the backlight data of the actual number of partitions and the image to be displayed to a preset display device for display; The actual number of partitions is smaller than the logical number of partitions, and the similarity between the display effect of the backlight data of the actual number of partitions output by the target backlight fusion model and the display effect of the backlight data of the logical number of partitions meets a preset similarity condition.

[0070] In addition, it is worth noting that the working process of a backlight fusion convolutional neural network image quality enhancement system provided in this embodiment is the same as the working process of the above-mentioned backlight fusion convolutional neural network image quality enhancement method, which will not be repeated here. For details, please refer to the working process of the above-mentioned backlight fusion convolutional neural network image quality enhancement method.

[0071] Figure 11 This is a schematic diagram of the structure of a terminal provided in an embodiment of the present application. The terminal may include: Memory 501 , processor 502 , and computer programs stored in the memory 501 and executable on the processor 502 .

[0072] When the processor 502 executes the program, the backlight fusion convolutional neural network image quality enhancement method provided in the above embodiment is implemented.

[0073] Furthermore, the terminal further includes: The communication interface 503 is used for communication between the memory 501 and the processor 502 .

[0074] The memory 501 is used to store computer programs that can be run on the processor 502 .

[0075] The memory 501 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0076] If memory 501, processor 502, and communication interface 503 are implemented independently, communication interface 503, memory 501, and processor 502 can be interconnected via a bus to facilitate communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, and control buses. For ease of illustration, the figure uses only one line, but this does not imply that there is only one bus or only one type of bus.

[0077] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can communicate with each other through an internal interface.

[0078] The processor 502 may be a central processing unit, or an application specific integrated circuit (ASIC), or configured to implement one or more integrated circuits of the embodiments of the present application.

[0079] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned backlight fusion convolutional neural network image quality enhancement method is implemented.

[0080] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the claims.

[0081] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or n embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0082] The logic and / or steps represented in the flowchart or otherwise described herein may be considered, for example, as a sequenced list of executable instructions for implementing the logical functions, and may be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can read and execute instructions from an instruction execution system, apparatus, or device).

[0083] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, n steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logical functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array, a field programmable gate array, etc.

[0084] It should be understood that the application of the present invention is not limited to the above examples. For those skilled in the art, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A backlight fusion convolutional neural network image quality enhancement method, characterized in that: The method comprises: Acquire an image to be displayed, and extract backlight data of a logical partition number corresponding to the image to be displayed; Inputting the backlight data of the logical partition number into a pre-trained target backlight fusion model to obtain the backlight data of the actual partition number output by the target backlight fusion model after fusing the backlight data of the logical partition number; Transmitting the backlight data of the actual number of partitions and the image to be displayed to a preset display device for display; The actual number of partitions is smaller than the logical number of partitions, and the similarity between the display effect of the backlight data of the actual number of partitions output by the target backlight fusion model and the display effect of the backlight data of the logical number of partitions meets a preset similarity condition.

2. The backlight fusion convolutional neural network image quality enhancement method according to claim 1, characterized in that: The training process of the pre-trained target backlight fusion model includes: A set of image sample data is selected from a pre-built training set as training samples, and a current image sample and a backlight sample of the corresponding logical partition number are obtained; Inputting the backlight samples of the logical partition number corresponding to the current image sample into a backlight fusion model constructed based on a convolutional neural network, and obtaining the backlight samples of the actual partition number output by the backlight fusion model; Performing display effect simulation on the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, respectively, to obtain a first simulation result corresponding to the backlight samples of the logical partition number and a second simulation result corresponding to the backlight samples of the actual partition number; Fine-tuning the parameters of the backlight fusion model according to the first simulation result and the second simulation result to obtain a current fine-tuned backlight fusion model; Determining whether all image sample data in the training set are selected as training samples; When all image sample data in the training set are selected as training samples, it is determined that one training of the backlight fusion model has been completed, and the current training number of the backlight fusion model is determined, and it is determined whether the current training number reaches a preset training number; When the current training times reaches the preset training times, the currently fine-tuned backlight fusion model is determined as the trained target backlight fusion model.

3. The backlight fusion convolutional neural network image quality enhancement method according to claim 2, characterized in that: After determining whether all image sample data in the training set are selected as training samples, the method further includes: When all the image sample data in the training set have not been selected as training samples, it is determined that one training of the backlight fusion model has not been completed, and the steps of selecting a group of image sample data from the pre-constructed training set as training samples, obtaining the current image sample and the backlight samples of the corresponding logical partition number, and subsequent steps are continued until all the image sample data in the training set have been selected as training samples; Furthermore, after determining whether the current number of training times reaches the preset number of training times, the method further includes: When the current number of training times does not reach the preset number of training times, the step of selecting a set of image sample data from the pre-constructed training set as training samples to obtain the current image sample and the backlight sample of its corresponding logical partition number and its subsequent steps are re-executed until the current number of training times reaches the preset number of training times.

4. The backlight fusion convolutional neural network image quality enhancement method according to claim 2, characterized in that: The display effect simulation is performed on the current image sample based on the backlight samples of the logical partition number and the backlight samples of the actual partition number, respectively, to obtain a first simulation result corresponding to the backlight samples of the logical partition number and a second simulation result corresponding to the backlight samples of the actual partition number, including: Performing diffusion processing on the backlight samples of the logical partition number to obtain a first diffusion result; Performing pixel compensation on the image sample according to the first diffusion result to obtain a first compensated image sample; Performing a display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters, and the backlight samples of the logical partition number, to obtain a first simulation result corresponding to the backlight samples of the logical partition number; Performing a display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters, and the backlight samples of the actual number of partitions, to obtain a second simulation result corresponding to the backlight samples of the actual number of partitions; Alternatively, diffusion processing is performed on the backlight samples of the logical partition number to obtain a first diffusion result, and diffusion processing is performed on the backlight samples of the actual partition number to obtain a second diffusion result; Performing pixel compensation on the image sample according to the first diffusion result to obtain a first compensated image sample, and performing pixel compensation on the image sample according to the second diffusion result to obtain a second compensated image sample; Performing a display effect simulation on the current image sample according to the pixel value corresponding to each pixel point on the first compensated image sample, the liquid crystal module parameters, and the backlight samples of the logical partition number, to obtain a first simulation result corresponding to the backlight samples of the logical partition number; The display effect of the current image sample is simulated according to the pixel values ​​corresponding to each pixel point on the second compensated image sample, the liquid crystal module parameters and the backlight samples of the actual partition number, and the second simulation result corresponding to the backlight samples of the actual partition number is obtained.

5. The backlight fusion convolutional neural network image quality enhancement method according to claim 4, characterized in that: The performing diffusion processing on the backlight samples of the logical partition number to obtain a first diffusion result includes: Performing diffusion processing on the backlight samples of the logical partition number by using anisotropic Gaussian blur diffusion method to obtain a first diffusion result; Furthermore, performing diffusion processing on the backlight samples of the actual number of partitions to obtain a second diffusion result includes: The backlight samples of the actual number of partitions are diffused using an isotropic Gaussian blur diffusion method to obtain a second diffusion result.

6. The backlight fusion convolutional neural network image quality enhancement method according to any one of claims 2 to 5, characterized in that: The convolutional neural network includes an average pooling layer, four convolutional layers and an activation function layer corresponding to each convolutional layer, and the first convolutional layer and its corresponding activation function layer serve as the input layer, the second convolutional layer and its corresponding activation function layer, the third convolutional layer and its corresponding activation function layer and the average pooling layer serve as the hidden layer, and the fourth convolutional layer and its corresponding activation function layer serve as the output layer.

7. The backlight fusion convolutional neural network image quality enhancement method according to claim 6, characterized in that: The calculation formula of the convolutional layer is: ； ； in, is the output channel index, Represents the convolutional layer The convolution result of the output channel at the (i, j) coordinate, i is the row index of the convolution layer input data or output data, j is the column index of the convolution layer input data or output data, Indicates the The offset value of each output channel, is the input channel index, is the convolution kernel size, u is the row index of the convolution kernel, v is the column index of the convolution kernel, is the input data of the c-th input channel of the convolutional layer at the coordinate (i+u, j+v), Represents the weight data of the c-th channel of the convolution kernel at the (u, v) coordinate; Furthermore, the calculation formula of the average pooling layer is: ； in, is the pooling result of the average pooling layer, The row index of the average pooling layer output data, is the column index of the average pooling layer output data, H is the height of the pooling window, W is the width of the pooling window, is the input data of the average pooling layer at the coordinate (x+h, y+w), x is the row index of the average pooling layer input data, and y is the column index of the average pooling layer input data.

8. A backlight fusion convolutional neural network image quality enhancement system, characterized in that: The system comprises: An acquisition module, configured to acquire an image to be displayed and extract backlight data of a logical partition number corresponding to the image to be displayed; A fusion module, configured to input the backlight data of the logical partition number into a pre-trained target backlight fusion model, and obtain the backlight data of the actual partition number output by the target backlight fusion model after fusing the backlight data of the logical partition number; A display module, configured to transmit the backlight data of the actual number of partitions and the image to be displayed to a preset display device for display; The actual number of partitions is smaller than the logical number of partitions, and the similarity between the display effect of the backlight data of the actual number of partitions output by the target backlight fusion model and the display effect of the backlight data of the logical number of partitions meets a preset similarity condition.

9. A terminal, characterized in that: include: A memory, a processor, and a backlight fusion convolutional neural network image quality enhancement program stored on the memory and runnable on the processor, wherein the backlight fusion convolutional neural network image quality enhancement program, when executed by the processor, implements the steps of the backlight fusion convolutional neural network image quality enhancement method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which can be executed to implement the steps of the backlight fusion convolutional neural network image quality enhancement method as described in any one of claims 1 to 7.

Images