Method for eliminating splicing seams between spliced display screens

Abstract

The invention belongs to the technical field of image display, and provides a method for eliminating splicing seams between spliced display screens, which comprises the following specific steps of: edge removal: preprocessing an image by using a Gaussian filter and morphological operation, identifying and removing edge noise by using a deep learning model, and retaining details and textures of the image; performing iterative content correction: adjusting the image by using a perspective transformation algorithm, and optimizing transformation parameters by using an algorithm based on gradient descent; quality evaluation and optimization, splicing and fusion are carried out; through accurate image correction and edge overlapping processing, smooth transition between spliced display screens can be ensured, visual fracture and discontinuity are reduced, the image quality of the whole display surface is improved, and the image looks more natural and has no obvious splicing seam or distortion at the splicing position; the correction technology can solve the problem of geometric distortion possibly existing in a single display screen, and the overlapping technology further ensures the consistency of images during cross-screen display.

Description

Technical Field

[0001] The invention belongs to the technical field of image display, and in particular is a method for eliminating seams between spliced ​​display screens. Background Art

[0002] At present, with the continuous advancement of electronic information technology, more and more computers are developing in the direction of miniaturization. At the same time as miniaturization, the computer's ability to process data has also been enhanced, and the type of data it can process has become more and more extensive. People can use small portable computers to complete large-scale data processing, storage, and collection work, which greatly facilitates people's daily work and life. For example, a laptop computer can be used to process stock information, map data, and image data, and display the corresponding content on the display screen based on the processing results. However, due to the miniaturization of computers, its display screen will inevitably be greatly reduced, and the reduced display screen will cause viewing difficulties for users. In particular, for the above data with high display quality requirements, the reduced display screen cannot meet the display quality requirements, so that the user cannot obtain the corresponding information in detail through the portable computer, which brings inconvenience to the user. For example, when a user downloads map data through a laptop computer, due to the small display screen, the user cannot clearly obtain the map with a large amount of information that he needs, such as a large map to determine the approximate location and a local detailed map showing the route. For another example, if a user processes stock data through a laptop computer, due to the small display screen, the user cannot clearly obtain the corresponding large amount of stock information, such as the overall market trend and individual stock details. In addition, as computers have become increasingly powerful in processing multimedia data such as videos, users can also use portable devices such as laptop computers to play videos. Although these portable devices have sufficient video data processing capabilities, due to the size limitations of their display screens, they cannot provide viewers with the visual enjoyment that a large screen can provide, thereby reducing the video data display capabilities of portable devices. In addition, in the process of using computers for engineering drawing and multi-window operations, the display screen also needs to be larger in size to facilitate actual use by users;

[0003] The prior art process through LCD fills liquid crystal capable of changing color and gray under the action of voltage in a small space, a plurality of small spaces form a display screen, the edge of the LCD is protected by a wrapping material to prevent leakage of the liquid crystal, and the width of the edge cannot be less than 5 mm in general; the CRT displays by means of electrons hitting the fluorescent screen on the inner surface of the display tube, the display tube is made of a glass shell with a certain thickness, and the middle is vacuumized, in order to ensure a certain strength, the glass shell needs to have a certain thickness, so the four edges of the CRT display tube cannot be reduced, and the minimum width is the thickness of the glass shell, which is generally 5 mm;

[0004] Therefore, due to the edge width of the CRT and the LCD respectively, a seam inevitably appears in the display screen splicing process. SUMMARY

[0005] In order to solve the above technical problems, the present application provides a method for eliminating the seam between spliced display screens to solve the problems mentioned in the background art.

[0006] A method for eliminating the seam between spliced display screens, comprising the following specific steps:

[0007] S1, edge removal: using a Gaussian filter and morphological operation to pre-process the image, and using a deep learning model to identify and remove edge noise while preserving the details and textures of the image;

[0008] S2, iterative content correction: gradually correcting the image, using a perspective transformation algorithm to adjust the image, including rotation, scaling and translation operations to restore the original form of the image, and using a gradient descent-based algorithm to optimize the transformation parameters;

[0009] S3, quality evaluation and optimization: evaluating the quality of the corrected image, if it meets the established quality standard, the processing is ended, if the image quality does not meet the standard, return to S2 for further iterative optimization;

[0010] S4, splicing and fusion: when both images have completed edge removal and content correction, splicing and fusion are performed, the edges of adjacent display screens are slightly overlapped during splicing, and a weighted average fusion method is used for the two images with overlapping parts, the weight of the overlapping area is calculated through an s-shaped curve, and then the fusion is performed to eliminate the seam.

[0011] Preferably, the edge removal specific steps include:

[0012] S101, using a Gaussian kernel to perform Gaussian blur processing on the original image through edge detection and corner detection technology;

[0013] S102. Based on the Gaussian blur, use the Sobel operator to calculate the derivatives in the horizontal and vertical directions, then calculate the magnitude of the gradient based on these derivatives, and determine the gradient direction of each pixel;

[0014] S103, using the pre-trained CNN model to perform a convolution operation on the image, applying the ReLU activation function after the convolution to increase nonlinearity, and judging whether the pixel belongs to the edge area based on the convolution and activation results, replacing the pixel determined to be the edge with the background color;

[0015] S104, combining the hierarchical features learned by the deep learning model with the information obtained from edge detection and corner detection to obtain accurate edge and key point positioning;

[0016] S105 , performing non-maximum suppression on the point with the local maximum gradient intensity, that is, the edge point, setting the gradient intensity of other points to zero, and connecting the broken edges through edge tracing to obtain a continuous edge line.

[0017] Preferably, the Sobel operator includes two 3x3 matrices, one for edge detection in the horizontal direction G x , and another one for vertical edge detection G y , then the horizontal gradient is expressed as:

[0018] G x =(-1×I x-1,y-1 )+(0×I x,y-1 )+(1×I x+1,y-1 )+(-2×I x-1,y )+(0×I x,y )+(2×I x+1,y )+(-1×I x-1,y+1 )+(0×I x,y+1 )+(1×I x+1,y+1 )

[0019] The vertical gradient is expressed as:

[0020] G y =(-1×I x-1,y-1 )+(-2×I x,y-1 )+(-1×I x+1,y-1 )+(0×I x-1,y )+(0×I x,y )+(0×I x+1,y )+(1×I x-1,y+1 )+(2×I x,y+1 )+(1×I x+1,y+1 )

[0021] Then the total gradient size is: G=|G x |+|G y|

[0022] The gradient direction is expressed as:

[0023] Where I represents the grayscale value of the image at the pixel point.

[0024] Preferably, the convolution operation is expressed as: C(x,y)=∑I(i,j)·K(xi,yj), where C(x,y) represents the convolution result, I(i,j) represents the pixel value of the image, and K(xi,yj) represents the convolution kernel function;

[0025] After convolution, the ReLU activation function is applied, and the output after activation is A(x,y)=max(0,C(x,y));

[0026] According to the results of convolution and activation, it is determined whether the pixel belongs to the edge area. The edge judgment is expressed as: Where E(x,y) represents the result of edge determination, and T is a preset threshold;

[0027] Replace the pixels judged as edges with the background color, expressed as: Where E(x,y) represents a binary edge detection image, E(x,y)=1 indicates that there is an edge at position (x,y), B represents the background value, which is used to mark the edge position in the edge detection image, and I(x,y) represents the pixel value of the original image at position (x,y).

[0028] Preferably, the gradient descent-based algorithm calculates the gradient of the loss function with respect to each parameter, i.e. Where L represents the loss function, θ i represents the i-th parameter; according to the calculated gradient and the preset learning rate, the learning rate is a hyperparameter that controls the step size, and the parameter value is updated, which is expressed as: Where η is the learning rate, and the above steps of calculating the gradient and updating the parameters are repeated until the preset number of iterations is reached or the gradient approaches zero.

[0029] Preferably, the weighted average fusion method generates a new fused image by performing a linear weighted average of the grayscale values ​​of the source image pixels, which is expressed as F(x, y) = w a A(x,y)+w b B(x,y), where w a and w b represents the weighting coefficient, which is used to control the contribution of each image in the fusion process, and w a +w b=1, F(x, y) represents the pixel value of the fused image, A(x, y) and B(x, y) represent the pixel values of two different images at the same position (x, y) respectively;

[0030] The S-shaped curve is used to calculate the weight, so that the weight in the overlapping area is smoothly transitioned from one image to another image, and the weight is assigned according to the position of the pixel distance from the center of the overlap, which is expressed as: Where k represents a parameter for controlling the slope of the curve, and x represents the position of the pixel relative to the center of the overlap. After determining the weight of each pixel, the weight is applied to the corresponding pixel to smoothly transition the two images and eliminate the seams.

[0031] Compared with the prior art, the present application has the following beneficial effects:

[0032] The present application can ensure smooth transition between the spliced display screens, reduce visual discontinuity and discontinuity, improve the image quality of the entire display surface, and make the image at the splicing position look more natural without obvious seams or distortion. The correction technology can solve the geometric distortion problem that may exist in a single display screen, and the overlap technology further ensures the consistency of the image when displaying across screens. The present method allows a certain degree of hardware tolerance, reduces the need for high-precision mechanical docking, thereby reducing production costs and improving overall structural stability. At the same time, the present method can adapt to display screens of different sizes and models, has high versatility and flexibility, and is suitable for a variety of different splicing application scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The present application is a method for eliminating seams between spliced display screens.

[0034] Figure 2 The present application is a method for eliminating seams between spliced display screens. DETAILED DESCRIPTION

[0035] The embodiments of the present application will be further described in detail below in conjunction with the drawings and examples. The following examples are used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0036] As shown in the accompanying Figure 1 to the accompanying Figure 2 As shown in the accompanying

[0037] Example 1: The present application provides a method for eliminating seams between spliced display screens, comprising the following specific steps:

[0038] S1, edge removal: using a Gaussian filter and morphological operation to pre-process the image, and using a deep learning model to identify and remove edge noise while preserving the details and texture of the image; the specific steps include:

[0039] S101. Perform Gaussian blur processing on the original image using a Gaussian kernel through edge detection and corner detection technology, thereby reducing the interference of noise in the image.

[0040] S102. Based on the Gaussian blur, use the Sobel operator to calculate the derivatives in the horizontal and vertical directions, then calculate the magnitude of the gradient based on these derivatives, and determine the gradient direction of each pixel;

[0041] The Sobel operator consists of two 3x3 matrices, one for horizontal edge detection G x , and another one for vertical edge detection G y , then the horizontal gradient is expressed as:

[0042] G x =(-1×I x-1,y-1 )+(0×I x,y-1 )+(1×I x+1,y-1 )+(-2×I x-1,y )+(0×I x,y )

[0043] +(2×I x+1,y )+(-1×I x-1,y+1 )+(0×I x,y+1 )+(1×I x+1,y+1 )

[0044] The vertical gradient is expressed as:

[0045] G y =(-1×I x-1,y-1 )+(-2×I x,y-1 )+(-1×I x+1,y-1 )+(0×I x-1,y )+(0×I x,y )

[0046] +(0×I x+1,y )+(1×I x-1,y+1 )+(2×I x,y+1 )+(1×I x+1,y+1 )

[0047] Then the total gradient size is: G=|G x |+|G y |

[0048] The gradient direction is expressed as:

[0049] Where I represents the grayscale value of the image at the pixel point.

[0050] S103, using the pre-trained CNN model to perform a convolution operation on the image, applying the ReLU activation function after the convolution to increase nonlinearity, and judging whether the pixel belongs to the edge area based on the convolution and activation results, replacing the pixel determined to be the edge with the background color;

[0051] The convolution operation is expressed as: C(x,y) = ∑I(i,j) K(xi,yj), where C(x,y) represents the convolution result, I(i,j) represents the pixel value of the image, and K(xi,yj) represents the convolution kernel function;

[0052] After convolution, the ReLU activation function is applied, and the output after activation is A(x,y)=max(0,C(x,y));

[0053] According to the results of convolution and activation, it is determined whether the pixel belongs to the edge area. The edge judgment is expressed as: Where E(x,y) represents the result of edge determination, and T is a preset threshold;

[0054] Replace the pixels judged as edges with the background color, expressed as: Where E(x,y) represents a binary edge detection image, E(x,y) = 1 indicates that there is an edge at position (x,y), B represents the background value, which is used to mark the edge position in the edge detection image, and I(x,y) represents the pixel value of the original image at position (x,y). That is, if an edge is detected at position (x,y) E(x,y) = 1, then the image I′ new The pixel value at the position (x, y) is set to the background value B. If no edge is detected, E(x, y) = 1, then the image I' new The pixel value at the position (x, y) remains the pixel value of the original image I.

[0055] S104, combining the hierarchical features learned by the deep learning model with the information obtained from edge detection and corner detection to obtain accurate edge and key point positioning;

[0056] S105: Non-maximum suppression is performed on the points with the local maximum gradient intensity, i.e., edge points. The gradient intensities of other points are set to zero. This removes redundant responses generated during edge detection, helps refine edges, and reduces false detections. Edge tracking is then used to connect broken edges, aiming to connect broken edge fragments to form a complete edge, thereby obtaining a continuous edge line. This further refines edge information and removes false positives, ultimately outputting clear edges and key point locations after noise removal.

[0057] S2, iterative content correction: gradually correct the image, using a perspective transformation algorithm to adjust the image, including rotation, scaling, and translation operations to restore the original image shape, and using a gradient descent-based algorithm to optimize the transformation parameters;

[0058] The gradient descent algorithm calculates the gradient of the loss function with respect to each parameter, i.e. Where L represents the loss function, θ i represents the i-th parameter; according to the calculated gradient and the preset learning rate, the learning rate is a hyperparameter that controls the step size, and the parameter value is updated, which is expressed as: Where η is the learning rate, and the above steps of calculating the gradient and updating the parameters are repeated until the preset number of iterations is reached or the gradient approaches zero.

[0059] S3, quality assessment and optimization: Perform quality assessment on the corrected image. If it meets the established quality standards, the process ends. If the image quality does not meet the standards, return to S2 to continue iterative optimization.

[0060] S4. Stitching and Fusion: After both images have completed edge removal and content correction, they are stitched and fused. During stitching, the edges of adjacent displays are allowed to overlap slightly. A weighted average fusion method is used for the two images with overlapping portions. The weight of the overlapping area is calculated using an S-shaped curve, and then the images are fused to eliminate the stitching seams. First, the width d of the overlapping area is determined. Then, based on the position of each pixel from the center of the overlapping area, its weight is calculated using an S-shaped function. This produces a weight distribution that changes slowly near the center of the overlap and rapidly at the edges, thereby achieving a smooth transition between images. After the weight calculation is completed, the pixel value of each source image is multiplied by the corresponding weight, and all weighted pixel values ​​are added together to obtain the pixel value of the fused image. After processing, the pixel values ​​near the stitching line will gradually transition from one image to the other, reducing the sense of abruptness and improving visual coherence.

[0061] The weighted average fusion method generates a new fused image by performing a linear weighted average of the grayscale values ​​of the source image pixels, which is expressed as F(x,y)=w a A(x,y)+w b B(x,y), where w a and w b represents the weighting coefficient, which is used to control the contribution of each image in the fusion process, and w a +w b =1, F(x,y) represents the pixel value of the fused image, A(x,y) and B(x,y) represent the pixel values ​​of the two different images at the same position (x,y);

[0062] The weights are calculated using an S-shaped curve so that in the overlapping area, the weights transition smoothly from one image to another. The weights are assigned according to the distance of the pixel from the center of the overlap, which can be expressed as: Here, k represents the parameter that controls the slope of the curve. When the k value is large, the weight changes more rapidly, meaning the fusion area is narrower; when the k value is small, the weight changes more slowly, and the fusion area is wider. x represents the position of the pixel relative to the center of the overlap. After determining the weight of each pixel, the weight is applied to the corresponding pixel to achieve a smooth transition between the two images and eliminate the stitching seam. Specifically, the overlapping area between the images is first identified and a weight value is calculated for each pixel using an S-shaped curve. Pixels closer to the center of the overlapping area receive higher weights, while those closer to the edge receive lower weights. Based on the weight of each pixel, the pixel values ​​of the source images are weighted averaged according to the weight. This ensures that the pixel values ​​in the center area are more likely to come from one image, while the pixel values ​​in the edge area smoothly transition to the other image. Ultimately, the fusion result is a seamless image, in which the pixel values ​​near the stitching line gradually transition from one image to the other, thus eliminating the stitching seam.

[0063] As can be seen from the above, through precise image correction and edge overlap processing, a smooth transition between spliced ​​display screens can be ensured, visual breaks and discontinuities can be reduced, the image quality of the entire display surface is improved, and the image looks more natural at the splicing point, without obvious seams or distortion; the correction technology can solve the geometric distortion problem that may exist in a single display screen, while the overlap technology further ensures the consistency of the image when displayed across screens; this method allows a certain degree of hardware tolerance, reduces the need for high-precision mechanical docking, thereby reducing production costs and improving overall structural stability; at the same time, this method can adapt to display screens of different sizes and models, has high versatility and flexibility, and is suitable for a variety of different splicing application scenarios.

[0064] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment (i.e., those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention) may not be described.

[0065] It will be appreciated that in the development of any actual embodiment, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but will, for those of ordinary skill having the benefit of this disclosure, be a routine undertaking of design, fabrication, and production without undue experimentation.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for eliminating seams between spliced ​​display screens, characterized in that: The specific steps include: S1. Edge removal: Use Gaussian filters and morphological operations to preprocess the image, and use a deep learning model to identify and remove edge noise, preserving the image details and texture; S2, iterative content correction: gradually correct the image, using a perspective transformation algorithm to adjust the image, including rotation, scaling, and translation operations to restore the original image shape, and using a gradient descent-based algorithm to optimize the transformation parameters; S3, quality assessment and optimization: Perform quality assessment on the corrected image. If it meets the established quality standards, the process ends. If the image quality does not meet the standards, return to S2 to continue iterative optimization. S4. Stitching and Fusion: After both images have completed edge removal and content correction, they are stitched and fused. During stitching, the edges of adjacent displays are allowed to overlap slightly. A weighted average fusion method is used for the two overlapping images. The weight of the overlapping area is calculated using an S-shaped curve, and then the images are fused to eliminate the stitching seams.

2. A method for eliminating seams between spliced ​​display screens as claimed in claim 1, characterized in that: The edge removal steps include: S101, performing Gaussian blur processing on the original image using a Gaussian kernel through edge detection and corner detection technology; S102. Based on the Gaussian blur, use the Sobel operator to calculate the derivatives in the horizontal and vertical directions, then calculate the magnitude of the gradient based on these derivatives, and determine the gradient direction of each pixel; S103, using the pre-trained CNN model to perform a convolution operation on the image, applying the ReLU activation function after the convolution to increase nonlinearity, and judging whether the pixel belongs to the edge area based on the convolution and activation results, replacing the pixel determined to be the edge with the background color; S104, combining the hierarchical features learned by the deep learning model with the information obtained from edge detection and corner detection to obtain accurate edge and key point positioning; S105 , performing non-maximum suppression on the point with the local maximum gradient intensity, that is, the edge point, setting the gradient intensity of other points to zero, and connecting the broken edges through edge tracing to obtain a continuous edge line.

3. A method for eliminating seams between spliced ​​display screens as claimed in claim 2, characterized in that: The Sobel operator includes two 3x3 matrices, one for horizontal edge detection G x , and another one for vertical edge detection G y , then the horizontal gradient is expressed as: G x =(-1×I x-1,y-1 )+(0×I x,y-1 )+(1×I x+1,y-1 )+(-2×I x-1,y )+(0×I x,y ) +(2×I x+1,y )+(-1×I x-1,y+1 )+(0×I x,y+1 )+(1×I x+1,y+1 ) The vertical gradient is expressed as: G y =(-1×I x-1,y-1 )+(-2×I x,y-1 )+(-1×I x+1,y-1 )+(0×I x-1,y )+(0×I x,y ) +(0×I x+1,y )+(1×I x-1,y+1 )+(2×I x,y+1 )+(1×I x+1,y+1 ) The total gradient size is: G=|G x |+|G y | The gradient direction is expressed as: Where I represents the grayscale value of the image at the pixel point.

4. The method for eliminating seams between spliced ​​display screens according to claim 2, wherein: The convolution operation is expressed as: C(x,y)=∑I(i,j)·K(xi,yj), where C(x,y) represents the convolution result, I(i,j) represents the pixel value of the image, and K(xi,yj) represents the convolution kernel function; After convolution, the ReLU activation function is applied, and the output after activation is A(x,y)=max(0,C(x,y)); According to the results of convolution and activation, it is determined whether the pixel belongs to the edge area. The edge judgment is expressed as: Where E(x,y) represents the result of edge determination, and T is a preset threshold; Replace the pixels judged as edges with the background color, expressed as: Where E(x,y) represents a binary edge detection image, E(x,y)=1 indicates that there is an edge at position (x,y), B represents the background value, which is used to mark the edge position in the edge detection image, and I(x,y) represents the pixel value of the original image at position (x,y).

5. The method for eliminating seams between spliced ​​display screens according to claim 1, wherein: The gradient descent-based algorithm calculates the gradient of the loss function with respect to each parameter, i.e. Where L represents the loss function, θ i represents the i-th parameter; according to the calculated gradient and the preset learning rate, the learning rate is a hyperparameter that controls the step size, and the parameter value is updated, which is expressed as: Where η is the learning rate, and the above steps of calculating the gradient and updating the parameters are repeated until the preset number of iterations is reached or the gradient approaches zero.

6. The method for eliminating seams between spliced ​​display screens according to claim 1, wherein: The weighted average fusion method generates a new fused image by performing a linear weighted average of the grayscale values ​​of the source image pixels, which is expressed as F(x, y) = w a A(x,y)+w b B(x,y), where w a and w b represents the weighting coefficient, which is used to control the contribution of each image in the fusion process, and w a +w b =1, F(x,y) represents the pixel value of the fused image, A(x,y) and B(x,y) represent the pixel values ​​of the two different images at the same position (x,y); The weights are calculated using an S-shaped curve so that in the overlapping area, the weights transition smoothly from one image to another. The weights are assigned according to the distance of the pixel from the center of the overlap, which can be expressed as: Where k represents the parameter that controls the slope of the curve, and x represents the position of the pixel relative to the overlapping center. After determining the weight of each pixel, the weight is applied to the corresponding pixel to make the transition between the two images smooth and eliminate the stitching.

Images