LED display screen image correction method

Abstract

The invention belongs to the field of image correction, and particularly relates to an LED display screen image correction method, which comprises the following steps of: 1, acquiring an image to be corrected; step 2, distortion correction preprocessing; step 3, distortion correction; step 4, extracting brightness features of the image; and 5, brightness correction: outputting a learnable weight by the multi-layer perception MLP model, and calculating the target brightness based on the learnable weight. The device is simple in structure and reasonable in design, has distortion correction and brightness correction functions, determines a variable zoom factor based on the size ratio between the real size of the image and the target size of the image, and corrects pixel coordinates according to the variable zoom factor, so that the correctness of the distorted image is improved; the target brightness is calculated based on the brightness characteristics and the learnable weight, and then brightness matching is performed, so that the brightness uniformity of the target image displayed by the LED display screen is improved, and a high-quality enhanced image of the dark image is obtained.

Description

Technical Field

[0001] The invention relates to the technical field of image correction, and in particular to a correction method for an image on an LED display screen. Background Art

[0002] In traditional medical imaging, security monitoring, intelligent machine vision, etc., there are functional requirements for images on LED display screens. Not only must the images not fail, but they must also ensure the human eye’s observation effect.

[0003] From acquisition to display, images on LED screens need to be calibrated according to pre-calibrated requirements. The first step is to overcome the size distortion caused by the accuracy and setting of the optical imaging system. The second step is to adapt to LED screens with different aspect ratios. The third step is to adjust the brightness of the image to avoid the image being too dark or having dark corners that affect the visual observation effect.

[0004] In order to prevent the image function from failing and potential safety hazards from occurring, it is necessary to perform correct and reasonable correction on the collected images. In the prior art, it is not convenient to perform correct and reasonable correction on the collected images. Therefore, we propose a correction method for LED display images to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for correcting an image on an LED display screen.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The method for correcting an image on an LED display screen comprises the following steps:

[0008] Step 1: Get the image to be corrected, adjust the resolution, color temperature, color saturation and color balance, preview the image after adjustment, and proceed to the next step if it meets the requirements;

[0009] Adjust the resolution: According to the size of the display, select a resolution that suits your needs to obtain clear images and texts;

[0010] Color temperature adjustment: adjust the cold and warm tones of the LED screen to meet the color requirements of the human eye for different scenes. By adjusting the color temperature parameters, the color tone and light effect of the LED screen can be changed;

[0011] Color saturation adjustment: adjust the color vividness by adjusting the color saturation parameters of the LED screen, so that the screen displays rich and full color effects;

[0012] Color balance adjustment: By adjusting the color balance parameters of the LED screen, it can display balanced and harmonious colors to avoid color distortion and excessive deformation;

[0013] Step 2: Distortion correction preprocessing:

[0014] Step 201, read the corner points of the image to be corrected, and obtain the target frame of the image to be corrected according to the corner points;

[0015] Step 202: extract features from the image to be corrected to obtain a target contour point set, where the target contour point set W = [w1, ..., w i ,...,w m ],w i represents the i-th contour point;

[0016] Step 3: Distortion Correction:

[0017] Step 301, calculate the scaling factor: the computer calculates the scaling factor according to the formula Calculate the scaling factor λ corresponding to the i-th contour point i , where d i-1 Indicates the minimum distance between the image point and the target box of the image to be corrected, d i-2 Indicates the distance between the image point and the i-th contour point w i distance;

[0018] Step 302: scaling the pixel points according to the scaling factor;

[0019] Step 303, repeating steps 301 to 302 for each contour point in the target contour point set, obtaining an image after distortion correction, and then proceeding to step 4;

[0020] Step 4: Extract the brightness features of the image:

[0021] Step 401, dividing the image after distortion correction into r second sub-images;

[0022] Step 402: extracting brightness features of the second sub-image;

[0023] Step 5: Brightness calibration;

[0024] Step 501: construct a multi-layer perception MLP model, input the brightness feature of each second sub-image into the multi-layer perception MLP model, and the multi-layer perception MLP model outputs a learnable weight ω;

[0025] Step 502: The computer calculates the target brightness θ according to the formula θ=θ·ω, where θ represents the average brightness of the second sub-image;

[0026] Step 503: Determine the brightness ratio of each second sub-image on each color channel according to the target brightness; perform brightness matching to complete brightness correction.

[0027] Preferably, it also includes size adaptation correction: the specific method of size adaptation correction is: scaling the image that has completed brightness correction according to the aspect ratio of the LED display screen.

[0028] Preferably, in step 201, the CornerNet model is used to read the corner points of the image to be corrected; in step 202, the HED edge detection algorithm is used to extract features of the image to be corrected.

[0029] Preferably, in S1, the resolution of a 14-inch display screen is 800*600; the resolution of a 17-inch display screen is 1024*768; and the resolution of a 19-inch display screen is 1280*1024;

[0030] The suitable resolution for a 20-inch screen is 1600*900; for a 22-inch screen, the resolution is 1680*1050; for a 24-inch display, the resolution is 1920*1080; for 27 inches and above, the resolution is 3840*2160.

[0031] Preferably, the specific method of scaling the pixel points according to the scaling factor in step 302 includes:

[0032] If d i-1 is the distance between the image point and the upper horizontal frame line of the target frame, which is equal to w i The pixels with the same horizontal coordinate and vertical coordinate greater than the vertical coordinate of the center point are scaled by the scaling factor λ i Zoom towards the horizontal frame line on the target frame;

[0033] If d i-1 is the distance between the image point and the lower horizontal frame line in the target frame, which is equal to w i The pixels with the same horizontal coordinate and whose vertical coordinate is smaller than the vertical coordinate of the center point are scaled by the scaling factor λ i Zoom towards the lower horizontal border of the target box;

[0034] If d i-1 is the distance between the image point and the left vertical frame line in the target frame, which is equal to w i The pixels with the same ordinate and whose abscissa is smaller than the abscissa of the center point are scaled by the scaling factor λ. i Zoom towards the left vertical border of the target box;

[0035] If d i-1 is the distance between the image point and the right vertical frame line in the target frame, which is equal to w iThe pixels with the same ordinate and whose abscissa is greater than the abscissa of the center point are scaled by the scaling factor λ. i Zoom toward the right vertical border of the target box.

[0036] Preferably, the specific method of extracting the brightness feature of the second sub-image in step 402 includes:

[0037] Step 4021: convert the second sub-image into HSV format and only retain the Value channel to obtain the Value channel image of the second sub-image;

[0038] Step 4022: Establish a feature extraction network;

[0039] Step 4023: input the Value channel image into the feature extraction network, and the feature extraction network outputs brightness features.

[0040] Preferably, the image point in step 301 is the intersection of a first line segment and a second line segment which are perpendicular to each other, wherein the first line segment passes through the i-th contour point, and the second line segment passes through the center point of the image to be corrected.

[0041] Preferably, in step 401, the specific method of dividing the distortion-corrected image into r second sub-images is: dividing the distortion-corrected image into n equal parts in the horizontal direction and the vertical direction respectively to obtain n second sub-images. 2 images, and then divide the images at the outermost edges into equal parts again.

[0042] Preferably, in step one, the image to be corrected is obtained, color temperature adjustment, color saturation adjustment and color balance adjustment are performed, and then anti-aliasing setting is performed. Anti-aliasing is an image processing technology used to reduce the jagged effect of the edge of the image and adjust the level of anti-aliasing. Increasing the anti-aliasing level can make the image edge smoother.

[0043] Preferably, in step one, the image to be corrected is obtained, color temperature adjustment, color saturation adjustment and color balance adjustment are performed, and then the refresh rate is adjusted. The refresh rate refers to the number of times the display updates the image per second. A higher refresh rate can provide a smoother image. The highest refresh rate is selected to obtain the best image quality and smoothness.

[0044] In the present invention, the LED display screen image correction method has the following beneficial effects:

[0045] 1. The invention has a simple structure, reasonable design, and is easy to implement and operate.

[0046] 2. In the distortion correction of the present invention, contour extraction and target frame extraction are used, and a variable scaling factor is determined based on the size ratio between the real size of the image and the target size of the image. The pixel coordinates are corrected according to the variable scaling factor, thereby improving the accuracy of the distorted image.

[0047] 3. In the brightness correction of the present invention, the image to be corrected is first divided into multiple second sub-images, and the brightness features of the second sub-images are respectively extracted. Based on the brightness features and learnable weights, the target brightness is calculated, and then the brightness is matched so that the brightness value of each second sub-image in the image to be corrected is equal to the target brightness, thereby improving the brightness uniformity of the target image displayed by the LED display screen and obtaining a high-quality enhanced image of the dark image.

[0048] In summary, the present invention has a simple structure and a reasonable design, and has distortion correction and brightness correction. A variable scaling factor is determined based on the size ratio between the real size of the image and the target size of the image, and pixel coordinates are corrected according to the variable scaling factor, thereby improving the accuracy of the distorted image; based on the brightness characteristics and the learnable weights, the target brightness is calculated, and then the brightness is matched, thereby improving the brightness uniformity of the target image displayed by the LED display screen, and obtaining a high-quality enhanced image of the dark image. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 The present invention is a flow chart of the method.

[0050] Figure 2 It is a schematic diagram of the target frame and the marked contour points of the image to be corrected of the present invention.

[0051] Figure 3 It is a schematic diagram of dividing the second sub-image according to the present invention. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0053] Embodiment 1

[0054] like Figure 1 As shown, the method for correcting an image on an LED display screen comprises the following steps:

[0055] Step 1: Obtain the image to be corrected;

[0056] In a possible embodiment, an image acquisition device is used to acquire the image to be corrected, and the image to be corrected is transmitted to a computer. The image acquisition device may be an industrial camera, such as an industrial 3lineRGB color camera, which is connected to the computer via a data cable; the industrial camera used is preferably an industrial camera with a photoelectric sensor, which can be used to acquire images.

[0057] Step 2: Distortion correction preprocessing:

[0058] Step 201: read the corner points of the image to be corrected, and obtain the target frame of the image to be corrected according to the corner points.

[0059] In actual use, the CornerNet model is used to read the corner points of the image to be corrected. The CornerNet model can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method of step 201.

[0060] The CornerNet model uses a single convolutional neural network to detect the target box of the image to be corrected as a pair of key points, namely the upper left corner and the lower right corner. In addition, the CornerNet model also introduces corner pooling as a new type of pooling layer to help the CornerNet model better locate corners.

[0061] The CornerNet detection process is as follows: (1) Input an image; (2) Use a special CNN structure to extract features; (3) Send the features to two detection networks, one for detecting the upper left corner of the object and the other for detecting the lower right corner of the object; (4) Map the corner points to the same space and regard the corner points with relatively close distances as belonging to the same object; (5) Draw the upper left corner and lower right corner points of the same object in the image to obtain the target box of the image to be corrected.

[0062] like Figure 2 As shown, the upper left corner point and the lower right corner point constitute the target frame of the image to be corrected, and the target frame of the image to be corrected includes an upper horizontal frame line a, a lower horizontal frame line c, a left vertical frame line d and a right vertical frame line b.

[0063] Step 202: extract features from the image to be corrected to obtain a target contour point set, where the target contour point set W = [w1, ..., w i ,...,w m ],w i Indicates the i-th contour point. In actual use, the edge contour features of the image to be corrected are extracted through the HED edge detection algorithm.

[0064] The HED edge detection algorithm is used to detect the image to be corrected and obtain the target contour point set of the image to be corrected. The HED edge detection algorithm uses the VGG-16 network model as the backbone network, and performs the output of multiple feature layers of the VGG-16 network model, including conv1_2, conv2_2, conv3_3, conv4_3, and conv5_3. The last pooling layer and the subsequent FC layer are removed, and a sid output is added after each convolution layer. The convolution kernel size of the side output layer is 1*1, and the final output is the fusion of multiple sid output features. The network weights are initialized using transfer learning, and the weight fusion layer is used to connect each layer.

[0065] Step 3: Distortion Correction:

[0066] Step 301, calculate the scaling factor: the computer calculates the scaling factor according to the formula Calculate the i-th contour point w i The corresponding scaling factor λ i , where d i-1 Indicates the minimum distance between the image point and the target box of the image to be corrected, d i-2 Indicates the distance between the image point and the i-th contour point w i distance.

[0067] In a possible embodiment, the image point is the intersection of a first line segment and a second line segment that are perpendicular to each other, wherein the first line segment passes through the i-th contour point, and the second line segment passes through the center point of the image to be corrected.

[0068] like Figure 2 As shown, point A is the contour point, point C is the coordinate point, e is the first line segment, f is the second line segment, and the image point is point B.

[0069] Scaling factor λ i It is the size ratio between the real size of the image and the target size of the image, where the minimum distance between the image point and the target box of the image to be corrected represents the real size of the image d i-1 , the distance between the image point and the i-th contour point w i The distance represents the image target size d i-2 Therefore, for different contour points w i For example, the scaling factor λ i is a variable scaling factor. The variable scaling factor is determined according to the size ratio between the real size of the image and the target size of the image. The pixel coordinates are corrected according to the variable scaling factor, thereby improving the correctness of the distorted image.

[0070] Since the image enlargement ratio is fixed and finite, the scaling factor λ can be calculated in advance iIn the subsequent steps, the scaling factors corresponding to different contour points can be predetermined and designed according to the above principles.

[0071] Step 302: Scale the pixels according to the scaling factor:

[0072] If d i-1 is the distance between the image point and the upper horizontal frame line of the target frame, which is equal to w i The pixels with the same horizontal coordinate and vertical coordinate greater than the vertical coordinate of the center point are scaled by the scaling factor λ i Zoom towards the horizontal frame line on the target frame;

[0073] If d i-1 is the distance between the image point and the lower horizontal frame line in the target frame, which is equal to w i The pixels with the same horizontal coordinate and whose vertical coordinate is smaller than the vertical coordinate of the center point are scaled by the scaling factor λ i Zoom towards the lower horizontal border of the target box;

[0074] If d i-1 is the distance between the image point and the left vertical frame line in the target frame, which is equal to w i The pixels with the same ordinate and whose abscissa is smaller than the abscissa of the center point are scaled by the scaling factor λ. i Zoom towards the left vertical border of the target box;

[0075] If d i-1 is the distance between the image point and the right vertical frame line in the target frame, which is equal to w i The pixels with the same ordinate and whose abscissa is greater than the abscissa of the center point are scaled by the scaling factor λ. i Zoom toward the right vertical border of the target box.

[0076] It should be noted that if the image to be corrected is skewed, that is, there is an angle between the central axis of the image to be corrected and the horizontal vertical line, the image to be corrected should be rotated first. If the angle is positive, the image to be corrected is rotated clockwise by the angle; if the angle is negative, the image to be corrected is rotated counterclockwise by the angle, which is convenient for subsequent operations.

[0077] In actual use, each pixel point located on the first line segment is scaled according to the scaling factor so that the distorted image size can be equal to the target size.

[0078] Step 303: Repeat steps 301 to 302 for each contour point in the target contour point set, exhaustively complete the correction of each contour point, obtain the image after the distortion correction is completed, and then proceed to step 4.

[0079] In distortion correction, contour extraction and target frame extraction are used to determine the variable scaling factor based on the size ratio between the real size of the image and the target size of the image. The pixel coordinates are corrected according to the variable scaling factor, thereby improving the accuracy of the distorted image.

[0080] Step 4: Extract the brightness features of the image:

[0081] Step 401: Divide the distortion-corrected image into r second sub-images.

[0082] Step 401 divides the distortion-corrected image into r second sub-images. The specific method is as follows: divide the distortion-corrected image into n equal parts in the horizontal direction and the vertical direction to obtain n sub-images. 2 images, and then divide the images at the outermost edges into equal parts again.

[0083] like Figure 3 As shown, the image is first divided into 4 equal parts in the horizontal and vertical directions to obtain 16 segmented images, and then the 4n-4 edge images are divided into 4 equal parts again to obtain 52 second sub-images in total. The second sub-images obtained by the division in step 401 can carry more edge dark information of the image.

[0084] The present application divides the image into equal parts, extracts brightness features from the multiple second sub-images after the equal divisions, inputs the brightness features of each second sub-image into a multi-layer perception MLP model, and the multi-layer perception MLP model outputs the target brightness. All area blocks in the image participate in the calculation of the target brightness to avoid the target brightness being excessively affected by a single area block, thereby reflecting its target brightness.

[0085] Step 402: extracting brightness features of the second sub-image;

[0086] Step 4021: convert the second sub-image into HSV format and only retain the Value channel to obtain the Value channel image of the second sub-image;

[0087] Step 4022: Establish a feature extraction network;

[0088] In actual use, the ResNet series network model is selected as the feature extraction network. The ResNet series network model and its variant network model have excellent performance for general image recognition tasks. In this embodiment, the ResNet18 network module is used, 18 represents the network depth, and the ResNet18 network module has 17 convolutional layers and 1 FC layer.

[0089] Step 4023: input the Value channel image into the feature extraction network, and the feature extraction network outputs brightness features.

[0090] The second sub-image is converted into HSV format, and only the Value channel is retained to obtain the Value channel image of the second sub-image. The HSV format includes three elements. From the perspective of psychology and vision, the color space points out that the color perception of the human eye mainly includes three elements: Hue channel, Saturation channel and Value channel, where the VHue channel represents color. Therefore, this scheme uses the Value channel image in HSV format as the feature map to realize the brightness feature extraction of the second sub-image.

[0091] Step 5: Brightness calibration;

[0092] Step 501: construct a multi-layer perception MLP model, input the brightness feature of each second sub-image into the multi-layer perception MLP model, and the multi-layer perception MLP model outputs a learnable weight ω;

[0093] Assume that there are m sample images, each sample image has r second sub-images, and the r second sub-images have r brightness features, then the input layer X∈R of the MLP model m×r , the output layer of the multi-layer perception MLP model outputs the probability weight ω.

[0094] Step 502: The computer calculates the target brightness θ according to the formula θ=θ·ω, where θ represents the average brightness of the second sub-image;

[0095] Step 503: Determine the brightness ratio of each second sub-image on each color channel according to the target brightness; perform brightness matching to complete brightness correction.

[0096] The average brightness of the second sub-image is calculated, and the target brightness is obtained according to the product of the average brightness θ and the learnable weight ω. Then, according to the target brightness and the initial brightness of the second sub-image, the ratio brightness of each second sub-image on each color channel is determined. Finally, the brightness of the second sub-image is adjusted according to the initial brightness of the second sub-image, thereby realizing the brightness correction of the entire image to be corrected, reducing the complexity and amount of calculation in the process of image brightness adjustment, and improving the adjustment efficiency and display effect of image brightness.

[0097] After completing the brightness matching, the brightness value of each second sub-image in the image to be corrected can be made equal to the target brightness value, thereby improving the brightness uniformity of the target image displayed by the LED display screen, and obtaining a high-quality enhanced image of the dark image, that is, improving the uniformity of the low grayscale image.

[0098] Step 502: Determine the brightness ratio of each second sub-image on each color channel according to the target brightness; perform brightness matching to complete brightness correction.

[0099] In brightness correction, the image to be corrected is first divided into multiple second sub-images, and the brightness features of the second sub-images are extracted respectively. Based on the brightness features and learnable weights, the target brightness is calculated, and then the brightness is matched so that the brightness value of each second sub-image in the image to be corrected is equal to the target brightness, thereby improving the brightness uniformity of the target image displayed by the LED display screen and obtaining a high-quality enhanced image of the dark image.

[0100] After the brightness correction is completed, the size adaptation correction is also included. The specific method of the size adaptation correction is: the image that has completed the brightness correction is proportionally scaled according to the aspect ratio of the LED display screen.

[0101] Embodiment 2

[0102] The difference from the first embodiment is that:

[0103] The method for correcting an image on an LED display screen comprises the following steps:

[0104] Step 1: Get the image to be corrected, adjust the resolution, color temperature, color saturation and color balance, preview the image after adjustment, and proceed to the next step if it meets the requirements;

[0105] Adjust the resolution: According to the size of the display, select a resolution that suits your needs to obtain clear images and texts;

[0106] Color temperature adjustment: adjust the cold and warm tones of the LED screen to meet the color requirements of the human eye for different scenes. By adjusting the color temperature parameters, the color tone and light effect of the LED screen can be changed;

[0107] Color saturation adjustment: adjust the color vividness by adjusting the color saturation parameters of the LED screen, so that the screen displays rich and full color effects;

[0108] Color balance adjustment: By adjusting the color balance parameters of the LED screen, it can display balanced and harmonious colors to avoid color distortion and excessive deformation;

[0109] Step 2: Distortion correction preprocessing:

[0110] Step 201, read the corner points of the image to be corrected, and obtain the target frame of the image to be corrected according to the corner points;

[0111] Step 202: extract features from the image to be corrected to obtain a target contour point set, where the target contour point set W = [w1, ..., w i ,...,w m ],w i represents the i-th contour point;

[0112] Step 3: Distortion Correction:

[0113] Step 301, calculate the scaling factor: the computer calculates the scaling factor according to the formula Calculate the scaling factor λ corresponding to the i-th contour point i , where d i-1 Indicates the minimum distance between the image point and the target box of the image to be corrected, d i-2 Indicates the distance between the image point and the i-th contour point w i distance;

[0114] Step 302: scaling the pixel points according to the scaling factor;

[0115] Step 303, repeating steps 301 to 302 for each contour point in the target contour point set, obtaining an image after distortion correction, and then proceeding to step 4;

[0116] Step 4: Extract the brightness features of the image:

[0117] Step 401, dividing the image after distortion correction into r second sub-images;

[0118] Step 402: extracting brightness features of the second sub-image;

[0119] Step 5: Brightness calibration;

[0120] Step 501: construct a multi-layer perception MLP model, input the brightness feature of each second sub-image into the multi-layer perception MLP model, and the multi-layer perception MLP model outputs a learnable weight ω;

[0121] Step 502: The computer calculates the Calculate the target brightness θ, where θ represents the average brightness of the second sub-image;

[0122] Step 503: Determine the brightness ratio of each second sub-image on each color channel according to the target brightness; perform brightness matching to complete brightness correction.

[0123] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for correcting an image on an LED display screen, characterized in that: The following steps are involved: Step 1: Get the image to be corrected, adjust the resolution, color temperature, color saturation and color balance, preview the image after adjustment, and proceed to the next step if it meets the requirements; Adjust the resolution: According to the size of the display, select a resolution that suits your needs to obtain clear images and texts; Color temperature adjustment: adjust the cold and warm tones of the LED screen to meet the color requirements of the human eye for different scenes. By adjusting the color temperature parameters, the color tone and light effect of the LED screen can be changed; Color saturation adjustment: adjust the color vividness by adjusting the color saturation parameters of the LED screen, so that the screen displays rich and full color effects; Color balance adjustment: By adjusting the color balance parameters of the LED screen, it can display balanced and harmonious colors to avoid color distortion and excessive deformation; Step 2: Distortion correction preprocessing: Step 201, read the corner points of the image to be corrected, and obtain the target frame of the image to be corrected according to the corner points; Step 202: extract features from the image to be corrected to obtain a target contour point set, where the target contour point set W = [w1, ..., w i ,...,w m ],w i represents the i-th contour point; Step 3: Distortion Correction: Step 301, calculate the scaling factor: the computer calculates the scaling factor according to the formula Calculate the scaling factor λ corresponding to the i-th contour point i , where d i-1 Indicates the minimum distance between the image point and the target box of the image to be corrected, d i-2 Indicates the distance between the image point and the i-th contour point w i distance; Step 302: scaling the pixel points according to the scaling factor; Step 303, repeating steps 301 to 302 for each contour point in the target contour point set, obtaining an image after distortion correction, and then proceeding to step 4; Step 4: Extract the brightness features of the image: Step 401, dividing the image after distortion correction into r second sub-images; Step 402: extracting brightness features of the second sub-image; Step 5: Brightness calibration; Step 501: construct a multi-layer perception MLP model, input the brightness feature of each second sub-image into the multi-layer perception MLP model, and the multi-layer perception MLP model outputs a learnable weight ω; Step 502: The computer calculates the Calculate the target brightness θ, where represents the average brightness of the second sub-image; Step 503: determining the matching brightness of each second sub-image on each color channel according to the target brightness; Perform brightness matching and complete brightness calibration.

2. The method for correcting an LED display screen image according to claim 1, characterized in that: It also includes size adaptation correction: the specific method of size adaptation correction is: scale the image that has completed brightness correction according to the aspect ratio of the LED display screen.

3. The method for correcting an LED display screen image according to claim 2, characterized in that: In step 201, the CornerNet model is used to read the corner points of the image to be corrected; in step 202, the HED edge detection algorithm is used to extract features of the image to be corrected.

4. The method for correcting an LED display screen image according to claim 3, characterized in that: In the S1, the resolution of the 14-inch display screen is 800*600; the resolution of the 17-inch display screen is 1024*768; the resolution of the 19-inch display screen is 1280*1024; The suitable resolution for a 20-inch screen is 1600*900; for a 22-inch screen, the resolution is 1680*1050; for a 24-inch display, the resolution is 1920*1080; for 27 inches and above, the resolution is 3840*2160.

5. The method for correcting an LED display screen image according to claim 4, characterized in that: The specific method of scaling the pixel points according to the scaling factor in step 302 includes: If d i-1 is the distance between the image point and the upper horizontal frame line of the target frame, which is equal to w i The pixels with the same horizontal coordinate and vertical coordinate greater than the vertical coordinate of the center point are scaled by the scaling factor λ i Zoom towards the horizontal frame line on the target frame; If d i-1 is the distance between the image point and the lower horizontal frame line in the target frame, which is equal to w i The pixels with the same horizontal coordinate and whose vertical coordinate is smaller than the vertical coordinate of the center point are scaled by the scaling factor λ i Zoom towards the lower horizontal border of the target box; If d i-1 is the distance between the image point and the left vertical frame line in the target frame, which is equal to w i The pixels with the same ordinate and whose abscissa is smaller than the abscissa of the center point are scaled by the scaling factor λ. i Zoom towards the left vertical border of the target box; If d i-1 is the distance between the image point and the right vertical frame line in the target frame, which is equal to w i The pixels with the same ordinate and whose abscissa is greater than the abscissa of the center point are scaled by the scaling factor λ. i Zoom toward the right vertical border of the target box.

6. The method for correcting an LED display screen image according to claim 5, characterized in that: The specific method of extracting the brightness feature of the second sub-image in step 402 includes: Step 4021: Convert the second sub-image into HSV format and only retain the Value channel to obtain the Value channel image of the second sub-image; Step 4022: Establish a feature extraction network; Step 4023: Input the Value channel image into the feature extraction network, and the feature extraction network outputs brightness features.

7. The method for correcting an LED display screen image according to claim 6, characterized in that: The image point in step 301 is the intersection of a first line segment and a second line segment that are perpendicular to each other, wherein the first line segment passes through the i-th contour point, and the second line segment passes through the center point of the image to be corrected.

8. The method for correcting an LED display screen image according to claim 7, characterized in that: In step 401, the specific method of dividing the distortion-corrected image into r second sub-images is as follows: the distortion-corrected image is divided into n equal parts in the horizontal direction and the vertical direction to obtain n second sub-images. 2 images, and then divide the images at the outermost edges into equal parts again.

9. The method for correcting an image on an LED display screen according to claim 8, characterized in that: In the step 1, the image to be corrected is obtained, and the color temperature, color saturation and color balance are adjusted, and then anti-aliasing settings are performed. Anti-aliasing is an image processing technology used to reduce the jagged effect of the edge of the image and adjust the level of anti-aliasing. Increasing the anti-aliasing level can make the image edge smoother.

10. The method for correcting an image on an LED display screen according to claim 9, characterized in that: In the step 1, the image to be corrected is obtained, and the color temperature, color saturation and color balance are adjusted, and then the refresh rate is adjusted. The refresh rate refers to the number of times the display updates the image per second. A higher refresh rate can provide a smoother image. The highest refresh rate is selected to obtain the best image quality and smoothness.

Images