Three-dimensional image synthesis method based on two-way vision

Abstract

The invention discloses a three-dimensional image synthesis method based on two-way vision, and relates to the technical field of vision and image processing, and the method comprises the following components: a two-way vision image collection module which is responsible for collecting image data of a left view angle and a right view angle, and carrying out the preprocessing of the collected image, the preprocessing comprises the steps of denoising, contrast enhancement, size adjustment and image alignment, and aims to improve the image quality and provide more accurate original data for subsequent matching and synthesis, a matching algorithm based on a deep neural network is adopted, and a large amount of image data is trained, so that the algorithm can learn a deep corresponding relationship between the images, and the image quality is improved. Therefore, higher-precision matching is realized, the problems of image distortion and fuzziness caused by matching errors are effectively reduced, a more reliable data basis is provided for subsequent three-dimensional image synthesis by the high-precision matching, and the definition and fidelity of the generated three-dimensional image are ensured.

Description

Technical Field

[0001] The present invention relates to the field of vision and image processing technology, and in particular to a stereoscopic image synthesis method based on dual-path vision. Background Art

[0002] With the continuous development of computer vision and image processing technology, two-way vision stereo image synthesis technology has become an important application means and is widely used in the fields of robot navigation, three-dimensional reconstruction, and virtual reality. Traditional two-way vision stereo image synthesis methods usually perform image matching based on feature point matching or phase matching algorithms to generate stereo images. However, these methods have great limitations in matching accuracy and stability and are easily affected by noise and image distortion, resulting in a decrease in matching accuracy.

[0003] Therefore, in order to improve the quality and accuracy of two-way vision stereo image synthesis, it is important to develop a stereo image synthesis method based on a deep neural network matching algorithm for two-way vision. The stereo image synthesis method based on two-way vision should have the following characteristics: according to the matching algorithm of the deep neural network, it can learn the deep feature relationship between images to achieve more accurate and stable matching, and it can be combined with the optimized synthesis algorithm to improve the quality and visual effect of the stereo image.

[0004] However, the traditional two-way vision stereo image synthesis methods on the market are often limited by the accuracy of matching algorithms. For scenes with less texture or poor lighting conditions, it is difficult to extract and match feature points. They are easily affected by noise and image distortion, resulting in inaccurate matching results and reduced accuracy. In the synthesis process, there is often a lack of optimization processing for image quality, resulting in distortion and blurring of the generated stereo images, which affects the visual effect and practical value of the stereo images. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the prior art and provide a stereoscopic image synthesis method based on dual-path vision. It can perform high-precision matching of dual-path vision images through the matching algorithm of deep neural network, construct a convolutional neural network model to accurately match the dual-path vision images, and use the optimized synthesis algorithm to more accurately fuse the image information of the two perspectives, reduce ghosting and blurring caused by perspective differences, thereby improving the clarity and overall visual effect of the stereoscopic image.

[0006] In order to solve the above technical problems, the present invention provides the following technical solution: a stereoscopic image synthesis method based on dual-path vision, the method comprising the following components:

[0007] Dual-channel visual image acquisition module: responsible for collecting image data from the left and right perspectives, and preprocessing the collected images. The preprocessing includes denoising, contrast enhancement, size adjustment, and image alignment steps, aiming to improve image quality and provide more accurate raw data for subsequent matching and synthesis.

[0008] High-precision matching algorithm module: It uses a matching algorithm based on a deep neural network to perform high-precision matching on pre-processed dual-channel visual images. The algorithm learns the correspondence between images by training a large amount of image data to achieve accurate matching. The module also includes optimization and correction functions for matching results to further improve matching accuracy and stability.

[0009] Stereoscopic image synthesis and post-processing module: Based on the matching results output by the high-precision matching algorithm module, this module fuses the image data of the left and right perspectives into a stereoscopic image. During the synthesis process, the synthesis algorithm is optimized to ensure that the generated stereoscopic image has high quality and realistic visual effects. In addition, this module also includes the function of post-processing the synthesized stereoscopic image, such as color correction and detail enhancement, to further improve the image quality;

[0010] User interaction and output module: This module is responsible for providing a user interaction interface, allowing users to view, edit and output the generated stereo images. Users can adjust and optimize the stereo images through this module to meet different application requirements. At the same time, this module also supports the output of the generated stereo images in multiple formats for subsequent application and processing.

[0011] Furthermore, the specific steps of the dual-channel visual image acquisition module are:

[0012] (1) Image acquisition: Use a dual-view camera system to simultaneously capture image data from both left and right perspectives, ensuring that the camera system is calibrated to reduce lens distortion and other possible errors;

[0013] (2) Noise removal: Apply the Gaussian filter denoising algorithm to reduce the noise in the image, smooth the image by calculating the statistical values ​​within the pixel neighborhood, and reduce random noise;

[0014] (3) Contrast enhancement: Use histogram equalization techniques to enhance the contrast of images. These techniques adjust the pixel intensity distribution of images to make the images clearer and the features more obvious.

[0015] (4) Image resizing: Adjust the image size as needed to match the input requirements of subsequent processing steps. Bilinear interpolation algorithms are usually used for image scaling.

[0016] (5) Image alignment: If there is a significant offset or rotation between the images from the two perspectives, image alignment is required to ensure that the images from the two perspectives are geometrically consistent.

[0017] Furthermore, the dual-channel visual image acquisition module smoothes the image and reduces Gaussian noise through a Gaussian filtering denoising algorithm. Gaussian filtering is usually used as the first step to denoise the original image. The value of each pixel is updated by calculating the weighted average of the pixels in its neighborhood. The weight is determined by a Gaussian function. The farther the pixel is from the center pixel, the smaller the weight. For each pixel (x, y), its filtered value G (x, y) is the weighted sum of the pixel value in its neighborhood and the Gaussian kernel, that is: w(i, j) is the weight of the Gaussian kernel, I(x+i, y+j) is the value of the pixel in the neighborhood, k is the size of the Gaussian kernel (usually an odd number), and the filtered image is smoother, the detail information is retained, and the noise is effectively suppressed, which is very beneficial for subsequent image processing.

[0018] Furthermore, the dual-channel visual image acquisition module stretches the image histogram through the histogram equalization technology to improve the contrast, which is used to improve the visual effect of the image and make the features in the image more obvious. By calculating the histogram of the image and adjusting the gray value of each pixel according to the histogram information, the distribution of the pixel gray value is more uniform. The operation mode and formula are as follows: First, the histogram Hk of the image is calculated: Hk = Σ i＝0 Σ j＝0 δIi,jk, and then calculate the cumulative distribution function CDFk: Finally, CDFk is used to map the image pixel values: Histogram equalization can effectively enhance the contrast of the image and make the dark and bright details in the image clearer. After equalization, the visual effect of the image is significantly improved, the feature information is more obvious, and it is conducive to subsequent image processing.

[0019] Furthermore, the dual-channel visual image acquisition module uses bilinear interpolation for image scaling and resampling. When the image size needs to be adjusted, bilinear interpolation is used to calculate the value of each pixel in the new image. The grayscale value of the target pixel is calculated by considering the grayscale values ​​of the four nearest neighboring pixels around the target pixel and their distances from the target pixel. For the pixel x′, y′ in the scaled image, its value fx′, y′ is the weighted sum of the four adjacent pixel values ​​in the original image. The algorithm formula is: fx′, y′≈1-α1-βfx1, y1+α1-βfx2, y1+1-αβfx1, y2+αβfx2, y2, where (x1, y1), (x2, y1), (x1, y2), (x2, y2) are the coordinates of the four nearest pixels to (x′, y′) in the original image, α and β are the interpolation coefficients of (x′, y′) in the horizontal and vertical directions relative to these four points, respectively. Bilinear interpolation can produce a smoother image scaling effect and avoid the jagged edges that may be produced by nearest neighbor interpolation. The resulting image is more visually natural and the detail information is retained, which is crucial for subsequent matching and feature extraction operations.

[0020] Furthermore, the high-precision matching algorithm module has the following specific steps:

[0021] (1) Feature extraction: Extract features from the preprocessed dual-path vision images. These features include edges, corners, and textures, which can represent important information in the image.

[0022] (2) Constructing a deep neural network model: Based on the extracted features, a convolutional neural network (CNN) model is constructed to learn the correspondence between images;

[0023] (3) Model training: Use a large amount of labeled image data to train the convolutional neural network model. Through the back-propagation algorithm, the model parameters are continuously adjusted so that the model can accurately match the dual-path visual images.

[0024] (4) Feature matching: The trained model is applied to the actual two-way visual image to match the extracted features. This is achieved by calculating the similarity between the features. Features with high similarity are considered to be matched.

[0025] (5) Generate disparity map: Generate a disparity map based on the result of feature matching. The disparity map represents the pixel offset between the left and right perspective images, which reflects the depth information of the scene.

[0026] (6) Optimization and correction: Post-process the disparity map to eliminate noise and fill holes to further improve the matching accuracy and stability.

[0027] Furthermore, the high-precision matching algorithm module constructs a convolutional neural network (CNN) model through the extracted features to learn the correspondence between images. CNN has powerful feature learning and expression capabilities and can extract hierarchical feature maps from images. The model construction process is: convolution operation is performed on the input feature map through the convolution kernel to extract local features. The algorithm formula is: Si, j = I*Ki, j = Σ m ∑ n Ii+m,j+nKm,n, I is the input feature map, K is the convolution kernel, S is the output feature map, the output of the convolution layer is transformed nonlinearly to increase the expressiveness of the model, the feature map is downsampled to reduce the amount of calculation and the number of parameters, the feature map is flattened into a one-dimensional vector, and linearly transformed through the weight matrix and bias term, and finally the output is obtained through the activation function.

[0028] Furthermore, the high-precision matching algorithm module uses a back-propagation algorithm to calculate the gradient of the loss function to the model parameters during the training process of building a convolutional neural network (CNN) model, and updates these parameters through an optimization algorithm. During the forward propagation process, the input data is output through the neural network, and the activation value and weight of each layer are calculated at the same time. For a certain layer (l) in the neural network, its output z l and activation value (a l ) can be expressed as: l =W l a l-1 +b l , a l =σz l , W l is the weight matrix, b l is the bias vector. The back propagation algorithm starts from the output layer and calculates the partial derivatives (gradients) of the loss function with respect to the weights and biases layer by layer. For the output layer L, its gradient ⊙ represents element-by-element multiplication, and its gradient δ l The gradient δ of the next layer can be l+1 Recursive calculation: δ l =W l+1T δ l+1 ⊙σ ′ z l , then, calculate the gradient of the loss function with respect to the weights and biases: Use the optimization algorithm to update the weights and biases according to the calculated gradients: Until the preset number of training rounds is reached or other stopping conditions are met (such as loss convergence), the back-propagation algorithm enables the model to self-adjust and optimize according to the labeled data, thereby continuously improving the matching accuracy. Through a large amount of training data and iterations, the model can learn the complex correspondence between images and perform well in practical applications, playing an important role in improving the matching accuracy and stability.

[0029] Furthermore, the stereo image synthesis and post-processing module generates a disparity map based on a high-precision matching algorithm, and the optimization synthesis algorithm assigns a weight to each pixel. The weight is usually calculated based on the reliability of the disparity map, the brightness, contrast or other image characteristics of the pixel. The purpose of weight assignment is to more accurately fuse the image information of the two perspectives. The optimization synthesis algorithm adopts a weighted average fusion strategy to fuse the image data of the left and right perspectives according to the weight and pixel value. In the weighted average fusion, the final value of each pixel is the weighted average of the pixels corresponding to the two perspectives. After the fusion is completed, the optimization synthesis algorithm also includes a color correction step to ensure that the generated stereo image is consistent in color. In addition, the algorithm also applies detail enhancement technology, such as Gaussian filtering, to further improve the detail performance of the image. The basic formula of the optimization synthesis algorithm is: stereo x,y=w L x,y·I L x,y+w R x,y·I R x,y,I stereo x, y are the pixel values ​​of the stereo image at coordinates (x, y), I L x, y and I R x, y are the pixel values ​​of the left and right view images at the corresponding coordinates, w L x, y and w R x, y are weight values ​​calculated based on the disparity map and other image features, satisfying w L x,y+w R x, y=1, the optimized synthesis algorithm can effectively fuse the image information of the left and right perspectives, reduce the ghosting and blurring caused by the difference in perspective, and thus improve the clarity of the stereoscopic image and the overall visual effect.

[0030] Furthermore, the method comprises the following steps:

[0031] (1) Image acquisition and preprocessing: Image data from the left and right perspectives are acquired and preprocessed, including denoising, contrast enhancement, size adjustment, and image alignment, to improve image quality.

[0032] (2) High-precision matching: A matching algorithm based on a deep neural network is used to perform high-precision matching on preprocessed images. The corresponding relationship between images is learned through training with a large amount of image data to ensure the accuracy of matching, and the matching results are optimized and corrected.

[0033] (3) Stereoscopic image synthesis and post-processing: The image data from the left and right perspectives are fused into a stereoscopic image. An optimized synthesis algorithm is used to ensure the image quality and visual effect. Post-processing, such as color correction and detail enhancement, is performed to further improve the image quality.

[0034] (4) User interaction and output: It provides an interactive interface that allows users to view, edit, and adjust the generated stereoscopic images, and supports output in multiple formats to facilitate subsequent application and processing.

[0035] Compared with the prior art, the stereoscopic image synthesis method based on dual-path vision has the following beneficial effects:

[0036] 1. The present invention adopts a matching algorithm based on a deep neural network. By training a large amount of image data, the algorithm can learn the deep correspondence between images, thereby achieving higher-precision matching and accurately determining the correspondence between the left and right perspective images, effectively reducing image distortion and blur problems caused by matching errors. This high-precision matching provides a more reliable data basis for subsequent stereo image synthesis, ensuring the clarity and realism of the generated stereo images.

[0037] 2. The invention further improves the quality of stereoscopic images by optimizing synthesis algorithms and post-processing technology. During the synthesis process, the use of advanced optimization algorithms ensures that the image data of the left and right perspectives can be more accurately integrated into a stereoscopic image. At the same time, the color correction and detail enhancement functions in the post-processing steps effectively improve the visual effect of the image, making it clearer and more realistic. This full-process optimization from matching to synthesis makes the generated stereoscopic images not only have higher accuracy, but also have better visual effects and practical value.

[0038] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 This is a flowchart of the stereo image synthesis method based on dual-path vision.

[0041] Figure 2 The flowchart of the stereo image synthesis method based on dual-path vision. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] In the first aspect, a stereoscopic image synthesis method based on dual-path vision includes the following components:

[0044] Dual-channel visual image acquisition module: responsible for collecting image data from the left and right perspectives, and preprocessing the collected images. The preprocessing includes denoising, contrast enhancement, size adjustment, and image alignment steps, aiming to improve image quality and provide more accurate raw data for subsequent matching and synthesis.

[0045] High-precision matching algorithm module: It uses a matching algorithm based on a deep neural network to perform high-precision matching on pre-processed dual-channel visual images. The algorithm learns the correspondence between images by training a large amount of image data to achieve accurate matching. The module also includes optimization and correction functions for matching results to further improve matching accuracy and stability.

[0046] Stereoscopic image synthesis and post-processing module: Based on the matching results output by the high-precision matching algorithm module, this module fuses the image data of the left and right perspectives into a stereoscopic image. During the synthesis process, the synthesis algorithm is optimized to ensure that the generated stereoscopic image has high quality and realistic visual effects. In addition, this module also includes the function of post-processing the synthesized stereoscopic image, such as color correction and detail enhancement, to further improve the image quality;

[0047] User interaction and output module: This module is responsible for providing a user interaction interface, allowing users to view, edit and output the generated stereo images. Users can adjust and optimize the stereo images through this module to meet different application requirements. At the same time, this module also supports the output of the generated stereo images in multiple formats for subsequent application and processing.

[0048] In a second aspect, a stereoscopic image synthesis method based on dual-path vision comprises the following steps:

[0049] (1) Image acquisition and preprocessing: Image data from the left and right perspectives are acquired and preprocessed, including denoising, contrast enhancement, size adjustment, and image alignment, to improve image quality.

[0050] (2) High-precision matching: A matching algorithm based on a deep neural network is used to perform high-precision matching on preprocessed images. The corresponding relationship between images is learned through training with a large amount of image data to ensure the accuracy of matching, and the matching results are optimized and corrected.

[0051] (3) Stereoscopic image synthesis and post-processing: The image data from the left and right perspectives are fused into a stereoscopic image. An optimized synthesis algorithm is used to ensure the image quality and visual effect. Post-processing, such as color correction and detail enhancement, is performed to further improve the image quality.

[0052] (4) User interaction and output: It provides an interactive interface that allows users to view, edit, and adjust the generated stereoscopic images, and supports output in multiple formats to facilitate subsequent application and processing.

[0053] Embodiment 1:

[0054] This embodiment provides a stereoscopic image synthesis method based on dual-path vision. By collecting dual-path vision images and performing preprocessing, the image quality is improved, laying the foundation for subsequent high-precision matching. The deep neural network technology is used to implement a high-precision matching algorithm, and the images of the left and right perspectives are accurately matched to generate a disparity map. Finally, a high-quality stereoscopic image is synthesized based on the matching results, and the user interaction and output module provide the user with viewing, editing and output functions. This process fully reflects the advantages of the present invention in the field of image processing and can generate clear and realistic stereoscopic images.

[0055] In the specific implementation, the system uses the equipped binocular camera to collect image data of the actual scene from the left and right perspectives, imports the collected image data into the image acquisition and preprocessing module, and uses Gaussian filtering: De-noising is performed to eliminate noise interference in the image, and contrast enhancement is performed on the image to highlight the details and features in the image and improve the image quality.

[0056] The preprocessed image data is input into the high-precision matching algorithm module, which uses a matching algorithm based on a deep neural network: Si,j = I*Ki,j = Σ m Σ n Ii+m,j+nKm,n, through the trained deep neural network model, the features of the images from the left and right perspectives are extracted and matched. During the matching process, the algorithm will automatically learn the correspondence between the images and generate a disparity map for subsequent stereo image synthesis.

[0057] According to the matching results and disparity map output by the high-precision matching algorithm module, the stereo image synthesis and post-processing module starts working, using the optimized synthesis algorithm: stereo x,y=w L x,y·I L x,y+w R x,y·I R x, y, the image data of the left and right perspectives are merged into a stereoscopic image. During the synthesis process, the parameters of the image brightness, color and contrast are adjusted to make the generated stereoscopic image more realistic and natural. The synthesized stereoscopic image is post-processed with detail enhancement and noise suppression to further improve the image quality.

[0058] Users can view the generated stereoscopic images and make necessary adjustments and optimizations through user interaction with the output module. The module provides an intuitive interface that allows users to scale, rotate and crop to meet different application requirements.

[0059] After completing the adjustment, users can choose to output the stereoscopic image in a variety of formats, such as JPEG, PNG or video format, for subsequent application and processing.

[0060] Embodiment 2:

[0061] The dual-path vision stereo image synthesis method in this embodiment particularly emphasizes that the convolutional neural network model performs nonlinear transformation during high-precision matching and stereo image synthesis, increases the expressive power of the model, downsamples the feature map, reduces the amount of calculation, the number of parameters, and ghosting and blurring caused by viewing angle differences, thereby improving the clarity of the stereo image and the overall visual effect.

[0062] The trained convolutional neural network model is deployed in the stereoscopic image synthesis and post-processing module, and the user interaction and output module fuses the image data from the left and right perspectives into a stereoscopic image. Users can adjust and optimize the stereoscopic image through this module to meet different application requirements.

[0063] Subsequently, the processed stereo image data is passed to the user interaction and output module. This transmission process ensures seamless connection between the two modules. When the user adjusts or optimizes the stereo image, these changes will be fed back to the stereo image synthesis and post-processing module in real time. This means that if the user performs operations such as changing the viewing angle or depth, the module will re-synthesize and post-process to generate a new stereo image.

[0064] In addition, the user can continuously optimize the effect of the stereoscopic image by viewing, adjusting and outputting multiple times. This cyclic optimization process enables the present invention to generate a stereoscopic image that better meets the needs of the user.

[0065] Embodiment 3:

[0066] The steps of the method based on Example 1 and Example 2 are as follows:

[0067] (1) Image acquisition and preprocessing: Image data from the left and right perspectives are acquired and preprocessed, including denoising, contrast enhancement, size adjustment, and image alignment, to improve image quality.

[0068] (2) High-precision matching: A matching algorithm based on a deep neural network is used to perform high-precision matching on preprocessed images. The corresponding relationship between images is learned through training with a large amount of image data to ensure the accuracy of matching, and the matching results are optimized and corrected.

[0069] (3) Stereoscopic image synthesis and post-processing: The image data from the left and right perspectives are fused into a stereoscopic image. An optimized synthesis algorithm is used to ensure the image quality and visual effect. Post-processing, such as color correction and detail enhancement, is performed to further improve the image quality.

[0070] (4) User interaction and output: It provides an interactive interface that allows users to view, edit, and adjust the generated stereoscopic images, and supports output in multiple formats to facilitate subsequent application and processing.

[0071] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. Stereoscopic image synthesis method based on dual-path vision, It is characterized in that The approach includes the following components: Dual-channel visual image acquisition module: responsible for collecting image data from the left and right perspectives, and preprocessing the collected images. The preprocessing includes denoising, contrast enhancement, size adjustment, and image alignment steps, aiming to improve image quality and provide more accurate raw data for subsequent matching and synthesis. High-precision matching algorithm module: It uses a matching algorithm based on a deep neural network to perform high-precision matching on pre-processed dual-channel visual images. The algorithm learns the correspondence between images by training a large amount of image data to achieve accurate matching. The module also includes optimization and correction functions for matching results to further improve matching accuracy and stability. Stereoscopic image synthesis and post-processing module: Based on the matching results output by the high-precision matching algorithm module, this module fuses the image data of the left and right perspectives into a stereoscopic image. During the synthesis process, the synthesis algorithm is optimized to ensure that the generated stereoscopic image has high quality and realistic visual effects. In addition, this module also includes the function of post-processing the synthesized stereoscopic image, such as color correction and detail enhancement, to further improve the image quality; User interaction and output module: This module is responsible for providing a user interaction interface, allowing users to view, edit and output the generated stereo images. Users can adjust and optimize the stereo images through this module to meet different application requirements. At the same time, this module also supports the output of the generated stereo images in multiple formats for subsequent application and processing.

2. The method for synthesizing stereoscopic images based on dual-path vision according to claim 1, It is characterized in that The specific steps of the dual-channel visual image acquisition module are as follows: (1) Image acquisition: Use a dual-view camera system to simultaneously capture image data from both left and right perspectives, ensuring that the camera system is calibrated to reduce lens distortion and other possible errors; (2) Noise removal: Apply the Gaussian filter denoising algorithm to reduce the noise in the image, smooth the image by calculating the statistical values ​​within the pixel neighborhood, and reduce random noise; (3) Contrast enhancement: Use histogram equalization techniques to enhance the contrast of images. These techniques adjust the pixel intensity distribution of images to make the images clearer and the features more obvious. (4) Image resizing: Adjust the image size as needed to match the input requirements of subsequent processing steps. Bilinear interpolation algorithms are usually used for image scaling. (5) Image alignment: If there is a significant offset or rotation between the images from the two perspectives, image alignment is required to ensure that the images from the two perspectives are geometrically consistent.

3. The method for synthesizing stereoscopic images based on dual-path vision according to claim 2, It is characterized in that The dual-channel visual image acquisition module smoothes the image and reduces Gaussian noise through a Gaussian filtering denoising algorithm. Gaussian filtering is usually used as the first step to denoise the original image. The value of each pixel is updated by calculating the weighted average of the pixels in its neighborhood. The weight is determined by a Gaussian function. The farther the pixel is from the center pixel, the smaller the weight. For each pixel (x, y), its filtered value G (x, y) is the weighted sum of the pixel value in its neighborhood and the Gaussian kernel, that is: w(i, j) is the weight of the Gaussian kernel, I(x+i, y+j) is the value of the pixel in the neighborhood, k is the size of the Gaussian kernel (usually an odd number), and the filtered image is smoother, the detail information is retained, and the noise is effectively suppressed, which is very beneficial for subsequent image processing steps.

4. The method for synthesizing stereoscopic images based on dual-path vision according to claim 2, It is characterized in that The dual-channel visual image acquisition module stretches the image histogram through the histogram equalization technology to improve the contrast, which is used to improve the visual effect of the image and make the features in the image more obvious. By calculating the histogram of the image and adjusting the gray value of each pixel according to the histogram information, the distribution of the pixel gray value is more uniform. Its operation mode and formula are as follows: First, the histogram H(k) of the image is calculated: Then calculate the cumulative distribution function CDF(k): Finally, CDF(k) is used to map the image pixel values: Histogram equalization can effectively enhance the contrast of the image and make the dark and bright details in the image clearer. After equalization, the visual effect of the image is significantly improved, the feature information is more obvious, and it is conducive to subsequent image processing.

5. The method for synthesizing stereoscopic images based on dual-path vision according to claim 2, It is characterized in that The dual-path visual image acquisition module uses bilinear interpolation for image scaling and resampling. When the image size needs to be adjusted, bilinear interpolation is used to calculate the value of each pixel in the new image. The grayscale value of the target pixel is calculated by considering the grayscale values ​​of the four nearest neighboring pixels around the target pixel and their distances from the target pixel. For a pixel (x′, y′) in the scaled image, its value f(x′, y′) is the weighted sum of the values ​​of the four adjacent pixels in the original image. The algorithm formula is: f(x′, y′)≈(1-α)(1-β)f(x′, y′)≈(1-α)(1-β) 1 ,y 1 )+α(1-β)f(x 2 ,y 1 )+91-α)βf(x 1 ,y 2 )+αβf(x 2 ,y 2 ), where (x 1 ,y 1 ), (x 2 ,y 1 ), (x 1 ,y 2 ), (x 2 ,y 2 ) are the coordinates of the four nearest pixels of (x′, y′) in the original image, α and β are the interpolation coefficients of (x′, y′) in the horizontal and vertical directions relative to these four points, respectively. Bilinear interpolation can produce a smoother image scaling effect, avoiding the jagged edges that may be produced by nearest neighbor interpolation. The resulting image is more visually natural and the detail information is preserved, which is crucial for subsequent matching and feature extraction operations.

6. The method for synthesizing stereoscopic images based on dual-path vision according to claim 1, It is characterized in that The specific steps of the high-precision matching algorithm module are as follows: (1) Feature extraction: Extract features from the preprocessed dual-path vision images. These features include edges, corners, and textures, which can represent important information in the image. (2) Constructing a deep neural network model: Based on the extracted features, a convolutional neural network (CNN) model is constructed to learn the correspondence between images; (3) Model training: Use a large amount of labeled image data to train the convolutional neural network model. Through the back-propagation algorithm, the model parameters are continuously adjusted so that the model can accurately match the dual-path visual images. (4) Feature matching: The trained model is applied to the actual two-way visual image to match the extracted features. This is achieved by calculating the similarity between the features. Features with high similarity are considered to be matched. (5) Generate disparity map: Generate a disparity map based on the result of feature matching. The disparity map represents the pixel offset between the left and right perspective images, which reflects the depth information of the scene. (6) Optimization and correction: Post-process the disparity map to eliminate noise and fill holes to further improve the matching accuracy and stability.

7. The method for synthesizing stereoscopic images based on dual-path vision according to claim 1, It is characterized in that The high-precision matching algorithm module constructs a convolutional neural network (CNN) model through the extracted features to learn the correspondence between images. CNN has powerful feature learning and expression capabilities and can extract hierarchical feature maps from images. The model construction process is: convolution operation is performed on the input feature map through the convolution kernel to extract local features. The algorithm formula is: S(i, j) = (I*K)(i, j) = ∑ m ∑ n I(i+m, j+n)K(m, n), I is the input feature map, K is the convolution kernel, S is the output feature map, and the output of the convolution layer is nonlinearly transformed to increase the expressiveness of the model. The feature map is downsampled to reduce the amount of calculation and the number of parameters. The feature map is flattened into a one-dimensional vector and linearly transformed through the weight matrix and bias term. Finally, the output is obtained through the activation function.

8. The method for synthesizing stereoscopic images based on dual-path vision according to claim 7, It is characterized in that The high-precision matching algorithm module uses the back-propagation algorithm to calculate the gradient of the loss function to the model parameters during the training process of building the convolutional neural network (CNN) model, and updates these parameters through the optimization algorithm. In the forward propagation process, the input data is output through the neural network, and the activation value and weight of each layer are calculated at the same time. For a certain layer (l) in the neural network, its output z l and activation value (a l ) can be expressed as: l =W l a l-1 +b l , a l =σ(z l ), W l is the weight matrix, b l is the bias vector. The back propagation algorithm starts from the output layer and calculates the partial derivatives (gradients) of the loss function with respect to the weights and biases layer by layer. For the output layer L, its gradient ⊙ represents element-by-element multiplication, and its gradient δ l The gradient δ of the next layer can be l+1 Recursive calculation: δ l =(W l+1 ) T δ l+1 ⊙σ′(z l ), then calculate the gradient of the loss function with respect to the weights and biases: Use the optimization algorithm to update the weights and biases according to the calculated gradients: Until the preset number of training rounds is reached or other stopping conditions are met (such as loss convergence), the back-propagation algorithm enables the model to self-adjust and optimize according to the labeled data, thereby continuously improving the matching accuracy. Through a large amount of training data and iterations, the model can learn the complex correspondence between images and perform well in practical applications, playing an important role in improving the matching accuracy and stability.

9. The method for synthesizing stereoscopic images based on dual-path vision according to claim 1, It is characterized in that The stereo image synthesis and post-processing module generates a disparity map based on a high-precision matching algorithm. The optimization synthesis algorithm assigns a weight to each pixel. The weight is usually calculated based on the reliability of the disparity map, the brightness, contrast or other image characteristics of the pixel. The purpose of weight assignment is to more accurately fuse the image information of the two perspectives. The optimization synthesis algorithm adopts a weighted average fusion strategy. According to the weight and pixel value, the image data of the left and right perspectives are fused. In the weighted average fusion, the final value of each pixel is the weighted average of the pixels corresponding to the two perspectives. After the fusion is completed, the optimization synthesis algorithm also includes a color correction step to ensure that the generated stereo image is consistent in color. In addition, the algorithm also applies detail enhancement technology, such as Gaussian filtering, to further improve the detail performance of the image. The basic formula of the optimization synthesis algorithm is: stereo (x, y) = w L (x, y)·I L (x,y)+w R (x, y)·I R (x,y),I stereo (x, y) is the pixel value of the stereo image at the coordinate (x, y), I L (x, y) and I R (x, y) are the pixel values ​​of the left and right view images at the corresponding coordinates, w L (x, y) and w R (x, y) is the weight value calculated based on the disparity map and other image characteristics, satisfying w L (x,y)+w R (x, y) = 1, the optimized synthesis algorithm can effectively fuse the image information of the left and right perspectives, reduce the ghosting and blurring caused by the difference in perspective, and thus improve the clarity of the stereoscopic image and the overall visual effect.

10. A calling number authentication method according to claim 1, It is characterized in that The method comprises the following steps: (1) Image acquisition and preprocessing: Image data from the left and right perspectives are acquired and preprocessed, including denoising, contrast enhancement, size adjustment, and image alignment, to improve image quality. (2) High-precision matching: A matching algorithm based on a deep neural network is used to perform high-precision matching on preprocessed images. The corresponding relationship between images is learned through training with a large amount of image data to ensure the accuracy of matching, and the matching results are optimized and corrected. (3) Stereoscopic image synthesis and post-processing: The image data from the left and right perspectives are fused into a stereoscopic image. An optimized synthesis algorithm is used to ensure the image quality and visual effect. Post-processing, such as color correction and detail enhancement, is performed to further improve the image quality. (4) User interaction and output: It provides an interactive interface that allows users to view, edit, and adjust the generated stereoscopic images, and supports output in multiple formats to facilitate subsequent application and processing.

Images