Infrared and visible image fusion method based on symbiotic statistical induced weighted total variation model

Abstract

The present application relates to the field of image processing, in particular to an infrared and visible light image fusion method based on a co-occurrence statistics induced weighted total variation model, the gray level of an infrared image is quantized, and a local gray co-occurrence matrix is constructed for each pixel point to calculate a co-occurrence probability; a spatial adaptive co-occurrence weight map capable of quantifying the texture flatness of a region is generated based on the probability; the weight map is introduced into a total variation regular term to construct a weighted total variation model, the model adaptively adjusts the smoothing strength of different regions of an image by using the weight map; the model is solved by using an alternating direction multiplier method to obtain a final fusion image. The present application realizes adaptive region differentiation of noise and edges, and effectively suppresses thermal noise of an infrared image. The present application realizes adaptive region differentiation of noise and edges by using co-occurrence statistics, effectively suppresses thermal noise of an infrared image, and retains the detail texture of a visible light image, effectively improving the visual quality and information integrity of a fusion image.

Description

Technical Field

[0001] This invention relates to the field of image processing, specifically to a method for fusing infrared and visible light images based on a co-occurrence statistically induced weighted total variation model. Background Technology

[0002] In applications such as security monitoring and driver assistance, infrared images can effectively capture thermal radiation targets, but they are sensitive to background thermal noise and lack texture details. Visible light images have rich scene details and texture information, but target recognition is low in low light or in bad weather. Image fusion technology aims to combine the advantages of both to generate an enhanced image that simultaneously highlights thermal targets and preserves detailed textures.

[0003] Currently, image fusion methods based on variational models have attracted widespread attention due to their mathematical rigor and high flexibility. Among them, the Total Variation (TV) model is commonly used for fusion tasks, achieving smoothing by minimizing image gradients while preserving edges. However, traditional TV models have significant shortcomings when applied to the fusion of infrared and visible light images: First, background thermal noise in infrared images has high-frequency characteristics similar to real edges, making it difficult for standard TV models to distinguish between noise and edges, easily preserving noise in the fusion result, leading to unclear images or artifacts, and limited ability to suppress thermal noise; Second, traditional TV models use globally uniform regularization parameters, which cannot dynamically adjust the smoothing intensity according to local image features, resulting in blurring of important details while suppressing noise, lacking spatial adaptability; Third, many related improvement methods construct weight maps based on gradient information, but gradient magnitudes cannot effectively distinguish between "strong noise" and "weak edges," resulting in insufficient accuracy of the weight map and affecting the fusion quality. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides an infrared and visible light image fusion method based on a co-occurrence statistics-induced weighted total variation model. By using co-occurrence statistics, the method adaptively distinguishes noise from edges, effectively suppresses thermal noise in infrared images, and preserves the detailed texture of visible light images, thereby significantly improving the visual quality and information integrity of the fused image.

[0005] The technical solution of this invention provides a method for fusion of infrared and visible light images based on a co-occurrence statistically induced weighted total variation model, comprising the following steps:

[0006] The system acquires the input infrared and visible light images, and performs grayscale quantization on the infrared image to obtain the quantized infrared image.

[0007] Based on the quantized infrared image, a local gray-level co-occurrence matrix is ​​constructed for each pixel, and the co-occurrence probability is calculated based on the local gray-level co-occurrence matrix.

[0008] The initial weight of each pixel is calculated based on the co-occurrence probability. The initial weight is then normalized and Gamma corrected to generate a spatially adaptive co-occurrence weight map.

[0009] An energy functional is constructed that includes an infrared image fidelity term, a visible light image fidelity term, and a total variation regularization term weighted by a spatially adaptive co-occurrence weight graph, to obtain a weighted total variation model;

[0010] The weighted total variation model is solved iteratively using the alternating direction multiplier method to obtain the final fused image.

[0011] In one optional implementation, grayscale quantization of the infrared image is performed, specifically including:

[0012] Quantize the original grayscale range of the infrared image to Each level, the quantized image any pixel pixel grayscale value Calculated using the following formula:

[0013]

[0014] in, This indicates the original infrared image at the pixel level. The pixel grayscale value at that location, This indicates the floor function.

[0015] In one optional implementation, the quantization level It is 128.

[0016] In one optional implementation, based on the quantized infrared image, a local gray-level co-occurrence matrix is ​​constructed for each pixel, and the co-occurrence probability is calculated based on the local gray-level co-occurrence matrix, specifically including:

[0017] For each pixel in the quantized infrared image, define a point centered on it. A local neighborhood window of a specific size; It is a positive integer greater than 1;

[0018] Within each local neighborhood window, calculate Gray-level co-occurrence matrix of size , where matrix elements This indicates that the grayscale value within this window is... The pixel and grayscale values The number of times that pixels appear together according to a preset spatial offset; This is the preset quantization level for grayscale quantization of infrared images;

[0019] gray-level co-occurrence matrix Normalization is performed to obtain the co-occurrence probability matrix corresponding to the local neighborhood window. , where matrix elements Indicates at pixel point Within a local neighborhood, grayscale pairs The probability of them appearing adjacent to each other in the target direction.

[0020] In an optional implementation, the initial weight of each pixel is calculated based on the co-occurrence probability, specifically including:

[0021] For the current pixel, form a gray-level pair between the gray-level value of the pixel and the gray-level values ​​of all other pixels in its local neighborhood window;

[0022] In the co-occurrence probability matrix In the middle, find the co-occurrence probability value corresponding to each gray-level pair;

[0023] The initial weight of a pixel is obtained by summing the co-occurrence probability values ​​of all grayscale pairs.

[0024] In one optional implementation, the correction coefficient for the Gamma correction process ranges from (0,1).

[0025] In one optional implementation, the energy functional is expressed as:

[0026]

[0027] in, For the desired fused image, The original infrared image, Visible light image, These are the weighting coefficients for the infrared and visible light fidelity terms, respectively. For regularization parameters, For spatial adaptive co-occurrence weight graph, For local neighborhood windows, To merge images gradient magnitude, This represents the Frobenius norm.

[0028] In an optional implementation, the weighted total variation model is iteratively solved using the alternating direction multiplier method to obtain the final fused image, specifically including:

[0029] Introducing auxiliary variables auxiliary variables As a constraint in the energy functional minimization problem;

[0030] Construct an augmented Lagrangian function for the constrained energy functional minimization problem;

[0031] Based on the augmented Lagrangian function, the solution of the minimization problem is decomposed into a problem concerning the fused image. Sub-problems and about auxiliary variables The two subproblems are solved iteratively and alternately until the convergence condition is met, and the final fused image is output. .

[0032] In one optional implementation, auxiliary variables The subproblem is solved by the weighted soft threshold shrinkage operator.

[0033] In an optional implementation, the threshold of the weighted soft threshold shrinkage operator is... ,in For spatial adaptive co-occurrence weight graph, To augment the penalty parameters of the Lagrange multiplier method.

[0034] As can be seen from the above technical solutions, this application has the following advantages:

[0035] (1) By constructing a spatial adaptive co-occurrence weight map based on the local gray-level co-occurrence matrix (GLCM), the texture "flatness" of each region of the image can be accurately quantified. High weights are given to flat background regions to enhance smoothness, and low weights are given to edge and detail regions to protect the structure, thereby achieving adaptive differentiation between noise and edges at the algorithm level.

[0036] (2) By introducing the co-occurrence weight graph into the total variation regularization term, a weighted total variation model is constructed, which enables the smoothing intensity to adapt to the weight space. This method can effectively suppress the background thermal noise of infrared images while preserving the texture details and target contours in visible light images to the greatest extent, significantly improving the visual clarity and information integrity of the fused images.

[0037] (3) The gray-level co-occurrence matrix is ​​used to statistically analyze the local texture mode. Compared with the traditional gradient method, it has stronger robustness to noise and local gray-level changes. Combined with the Gamma correction strategy, the dynamic range and distribution of the weight map are further optimized, ensuring the accuracy and stability of the weight map in complex scenes.

[0038] (4) By using the alternating direction multiplier method (ADMM) to solve the weighted variational model, the complex optimization problem is decomposed into multiple subproblems that are easy to solve. Furthermore, the fast Fourier transform (FFT) is used to accelerate the calculation, ensuring that the algorithm has good computational efficiency while maintaining high accuracy, and meeting the real-time requirements of practical applications. Attached Figure Description

[0039] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the process for an infrared and visible light image fusion method based on a symbiotic statistical induced weighted total variation model, provided as an embodiment of the present invention. Detailed Implementation

[0041] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0042] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0043] The key terms used in this invention will be explained below.

[0044] GLCM (Gray Level Co-occurrence Matrix): A statistical method for describing texture by studying the spatial correlation characteristics of gray levels.

[0045] TV Regularization (Total Variation Regularization): A commonly used image smoothing and denoising model with good edge preservation properties.

[0046] ADMM: Alternating Direction Multiplier Method, a computational framework for solving constrained optimization problems.

[0047] Figure 1 This is a schematic flowchart of an infrared and visible light image fusion method based on a co-occurrence statistically induced weighted total variation model, provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps.

[0048] S1: Acquire the input infrared image and visible light image, and perform grayscale quantization on the infrared image to obtain the quantized infrared image.

[0049] S2, based on the quantized infrared image, constructs a local gray-level co-occurrence matrix for each pixel, and calculates the co-occurrence probability based on the local gray-level co-occurrence matrix.

[0050] S3 calculates the initial weight of each pixel based on the co-occurrence probability, normalizes the initial weight and performs Gamma correction, and generates a spatially adaptive co-occurrence weight map.

[0051] S4. Construct an energy functional that includes an infrared image fidelity term, a visible light image fidelity term, and a total variation regularization term weighted by a spatially adaptive co-occurrence weight graph to obtain a weighted total variation model.

[0052] S5. The weighted total variation model is solved iteratively using the alternating direction multiplier method to obtain the final fused image.

[0053] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the specific implementation process of this embodiment, another embodiment of the infrared and visible light image fusion method based on the co-occurrence statistical induced weighted total variation model is provided, which includes the following steps.

[0054] S100: Acquire the original image and perform grayscale quantization on the infrared image in the original image.

[0055] In this embodiment, the input image is a registered infrared image of the same scene. With visible light images The images are sourced from a compatible infrared thermal imager and visible light camera simultaneous acquisition system, ensuring that the two images are precisely aligned in space and are the same size.

[0056] To facilitate subsequent processing, the infrared image is first... With visible light images The pixel values ​​are normalized to a grayscale range of [0, 255]. If the input is a color visible light image, it is first converted to its grayscale image. This normalization process provides a consistent numerical benchmark for subsequent quantization and numerical calculations.

[0057] To reduce the grayscale resolution of the infrared image, thereby reducing data sparsity and enhancing statistical stability during subsequent gray-level co-occurrence matrix (GLCM) statistics, and to effectively aggregate background thermal noise patterns, the original infrared image is first subjected to grayscale quantization to quantize the original grayscale range [0, 255] into... Each level. This embodiment selects a quantization level. This value, while retaining sufficient grayscale levels to distinguish different thermal radiation regions, can also aggregate certain noise patterns through subsequent statistical analysis, compared to... More robust than Less detail is lost.

[0058] infrared images Each pixel in Its original grayscale value Mapped to quantized grayscale values The mapping formula is:

[0059]

[0060] in, This indicates a floor function (rounding down). This mapping divides a continuous grayscale range into... The infrared image is quantized by dividing it into equal-width quantization intervals and mapping each pixel value to the integer representative value of its interval. After traversing all pixels of the infrared image, the quantized infrared image is obtained. Its size is the same as the original infrared image, but the grayscale range is changed from the original 0-255 to 0-127. Through the above quantization, the subtle grayscale fluctuations caused by sensor noise or weak thermal fluctuations in the original infrared image are suppressed, and the grayscale values ​​of similar thermal radiation areas (such as the sky and road surface) tend to be consistent. On the one hand, this reduces the dimensionality and computational complexity of subsequent GLCM construction, and on the other hand, it makes the background thermal noise more statistically easier to identify as a common pattern.

[0061] S200, Construction of Local Gray-Level Co-occurrence Matrix (GLCM) and Calculation of Co-occurrence Probability.

[0062] This step is based on the quantized infrared image, constructs a local gray-level co-occurrence matrix for each pixel, and calculates the co-occurrence probability based on the local gray-level co-occurrence matrix, specifically including the following steps S2001 to S2003.

[0063] S2001, for each pixel in the quantized infrared image, define a definition centered on... A local neighborhood window of a specific size; It is a positive integer greater than 1.

[0064] For quantized infrared images Each pixel in Define a space of size centered on it. Square sliding window In this embodiment, preferably... Specifically, a window size of 5×5 or 7×7 is used. This size choice is used to balance the locality and robustness of texture statistics: a window that is too small may lead to insufficient statistics and sensitivity to noise; a window that is too large may blur local details and weaken the distinguishability at edges. The window uses symmetrical padding at the image boundaries to maintain consistent size.

[0065] S2002, within each local neighborhood window, calculate Gray-level co-occurrence matrix of size , where matrix elements This indicates that the grayscale value within this window is... The pixel and grayscale values The number of times that pixels appear together according to a preset spatial offset; This is the preset quantization level for grayscale quantization of infrared images.

[0066] In each local neighborhood window Inside, construct a space of size Gray-level co-occurrence matrix ,in This refers to the quantization level. It's important to understand that the grayscale range of the quantized image is... The Gray-Level Co-occurrence Matrix (GLCM) is a statistical representation of all possible gray-level pairs. A square matrix of occurrences, with its index... and The range of values ​​is Therefore, the size of the matrix must be .

[0067] First, spatial relationships are defined. In this embodiment, the adjacency of pixel pairs in the horizontal rightward (0° direction) is considered, i.e., the spatial offset is set to... This can capture the most common horizontal texture structures in an image. In other embodiments, a neighborhood with a radius of 1 can be taken, or multiple directions (such as 0°, 45°, 90°, 135°) can be considered simultaneously, and the results of each direction can be averaged or the maximum value can be taken to enhance rotation invariance.

[0068] At each pixel position within the window As a reference point, based on the preset spatial offset Determine the positions of its adjacent pixels The grayscale value of the reference pixel. grayscale values ​​of adjacent pixels This constitutes a statistical unit. For each such statistical unit, in the gray-level co-occurrence matrix... In the middle, the first line, number The value of each element in a column is incremented by 1. Specifically, the matrix elements... This indicates that the grayscale value is within the local neighborhood window. The pixel and grayscale values The number of times that pixels co-occur according to a preset spatial offset is represented as:

[0069]

[0070] in: The quantized grayscale value. This is an indicator function that takes the value 1 when the condition is met, and 0 otherwise.

[0071] By iterating through all valid pixel pairs within the window, a 128×128 counting matrix is ​​finally obtained. This matrix represents the spatial co-occurrence frequency of all gray-level pairs in a specified direction within a local region.

[0072] S2003, regarding the gray-level co-occurrence matrix Normalization is performed to obtain the co-occurrence probability matrix corresponding to the local neighborhood window. , where matrix elements Indicates at pixel point Within a local neighborhood, grayscale pairs The probability of them appearing adjacent to each other in the target direction.

[0073] To eliminate the influence of window size on the absolute count value and facilitate subsequent calculations, the gray-level co-occurrence matrix is... After normalization, the local co-occurrence probability matrix is ​​obtained. Matrix elements The expression is:

[0074]

[0075] This normalization operation will normalize each matrix element. Divide by the sum of all elements in the matrix, and the matrix elements Indicates at pixel point Within a local neighborhood, grayscale pairs Based on the probability of adjacent occurrences in the target direction, satisfying .

[0076] In flat, uniform background regions, the gray values ​​of the central pixel are the same or very close to those of its surrounding pixels, resulting in significantly higher values ​​for elements along the main diagonal and in the vicinity of the co-occurrence probability matrix. Conversely, at edges or noise points, the gray values ​​of the central pixel often differ significantly from their neighboring pixels, forming "rare" gray value combinations, which correspond to very small values ​​in the probability matrix. Therefore, the distribution pattern of the probability matrix directly and quantitatively reflects the texture characteristics (flatness / complexity) of the local region where the current pixel is located, providing accurate statistical basis for the next step of generating an adaptive spatial weight map.

[0077] S300 generates a spatially adaptive co-occurrence weighted graph.

[0078] This step calculates the initial weight of each pixel based on the co-occurrence probability, normalizes the initial weight and performs Gamma correction, and generates a spatially adaptive co-occurrence weight map, which is to calculate the flatness weight of each pixel.

[0079] First, the initial weight of each pixel is calculated based on the co-occurrence probability. Specifically, this includes: for the current pixel, forming a gray-level pair between the pixel's gray value and the gray values ​​of all other pixels within its local neighborhood window; and then, in the co-occurrence probability matrix... In the process, the co-occurrence probability value corresponding to each grayscale pair is found; the co-occurrence probability values ​​corresponding to all grayscale pairs are summed to obtain the initial weight of that pixel. The lookup operation is as follows: for the center pixel... grayscale value With another pixel in the neighborhood grayscale value Directly index the pre-computed co-occurrence probability matrix Read its first line, number Column element values This value represents this specific grayscale pair. Probability estimate of occurrence within the current window.

[0080] Specifically, for the center pixel Its initial weight Defined as the sum (or average) co-occurrence probability of gray-level pairs formed by the pixel and its neighboring pixels, this reflects the "commonness" of the texture pattern in which the pixel is located, and is expressed as:

[0081]

[0082] like Against a flat background, the surrounding pixel values With center pixel value The same or similar items, and such pairings exist extensively within the window, leading to... The value is very large, therefore The value is large. If... Points located at the edge or in noise areas constitute "rare events" within the window, and their corresponding grayscale pairs are... The value is very small, therefore The value is small.

[0083] Because infrared images contain thermal noise in the background, direct calculation... In the background region, the values ​​are too small to provide sufficient regularization constraints. Gamma correction is introduced for nonlinear mapping.

[0084] First of all Normalization to Interval:

[0085]

[0086] Then, Gamma correction is applied to obtain the final weights. :

[0087]

[0088] in, As the correction factor, this embodiment takes... .when At this time, the transformation has the effect of "brightening dark areas". It can significantly increase the weight value of the background area (which originally has a small weight), so that it can trigger the smoothing and denoising mechanism; at the same time, it maintains the low weight characteristics of the edge area (with a weight close to 0), thus achieving a balance between denoising and edge preservation.

[0089] S400, construct a weighted total variation model.

[0090] This step constructs an energy functional that includes infrared image fidelity terms, visible light image fidelity terms, and a total variational regularization term weighted by a spatially adaptive co-occurrence weight graph, resulting in a weighted total variational model. Specifically, the expression for the energy functional is:

[0091]

[0092] The first term is the infrared image fidelity term, used to constrain the fused image to maintain consistency with the infrared image in overall grayscale distribution. Its expression is: ,in, For the desired fused image, The original infrared image, This is the weighting coefficient for this item. This represents the Frobenius norm.

[0093] The second term is the visible light image fidelity term, used to constrain the fused image to maintain consistency with the visible light image in terms of detail texture. Its expression is: ,in The input is a visible light image. This is the weighting coefficient for this item.

[0094] The third term is the weighted total variation regularization term, used to preserve edge structure while smoothing the image; its expression is: ,in For regularization parameters, For spatial adaptive co-occurrence weight graph, For local neighborhood windows, To merge images The gradient magnitude.

[0095] In the weighted total variation regularization term, the weight graph The smoothing intensity of the total variational regularization term is modulated spatially adaptively: a stronger smoothing constraint is applied in regions with larger weight values ​​to suppress noise; a weaker smoothing constraint is applied in regions with smaller weight values ​​to preserve edge and texture details. That is, in the background region, Larger gradients (after Gamma correction) force the model to minimize the gradient. This means performing strong noise reduction and smoothing; in the edge regions, The regularization term fails, and the model allows gradients to exist, thus perfectly preserving edge details.

[0096] S500, based on ADMM numerical solution.

[0097] This embodiment employs the alternating direction multiplier method to iteratively solve the weighted total variation model, obtaining the final fused image. Specifically, since the regularization term contains non-differentiable terms... The norm is solved using the ADMM method, which introduces auxiliary variables.

[0098] S5001, introducing auxiliary variables auxiliary variables As a constraint on the problem of minimizing energy functionals.

[0099] By introducing auxiliary variables, the minimization problem of the energy functional is transformed into the following optimization problem with equality constraints:

[0100]

[0101] S5002, Construct the augmented Lagrangian function for the constrained energy functional minimization problem.

[0102] To solve the above constrained optimization problem, we construct its augmented Lagrangian function, expressed as:

[0103]

[0104] in, Let Lagrange multiplier vectors be used. This is the penalty parameter.

[0105] S5003, based on the augmented Lagrangian function, decomposes the solution of the minimization problem into a problem concerning the fused image. Sub-problems and about auxiliary variables The two subproblems are solved iteratively and alternately until the convergence condition is met, and the final fused image is output. .

[0106] An alternating direction minimization strategy is used to update the variables sequentially for the augmented Lagrangian function. This continues until the convergence condition is met. In the ADMM... In this iteration, the following updates are performed sequentially.

[0107] (1) Update the merged image

[0108] fixed The following linear system is obtained by solving: :

[0109]

[0110] in, For the Laplace operator, As a divergence operator, this equation can be solved efficiently using the Fast Fourier Transform.

[0111] (2) Update auxiliary variables

[0112] In this embodiment, auxiliary variables The subproblem is solved using a weighted soft thresholding shrinkage operator, where the threshold is... ,in For spatial adaptive co-occurrence weight graph, To augment the penalty parameters of the Lagrange multiplier method.

[0113] Specifically, fixed and Updated using the following weighted soft threshold shrinkage operator :

[0114]

[0115] This operator achieves gradient shrinkage under weight modulation, and has both denoising and edge preservation functions.

[0116] (3) Update the Lagrange multipliers

[0117] Update the Lagrange multipliers based on the current variable values, expressed as:

[0118]

[0119] Repeat the above iterations until the images are merged in two consecutive iterations. The change is less than a preset threshold, or the maximum number of iterations is reached; the result obtained at final convergence will be... This is output as the fused image.

[0120] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for fusing infrared and visible light images based on a co-occurrence statistically induced weighted total variation model, characterized in that, Includes the following steps: The system acquires the input infrared and visible light images, and performs grayscale quantization on the infrared image to obtain the quantized infrared image. Based on the quantized infrared image, a local gray-level co-occurrence matrix is ​​constructed for each pixel, and the co-occurrence probability is calculated based on the local gray-level co-occurrence matrix. The initial weight of each pixel is calculated based on the co-occurrence probability. The initial weight is then normalized and Gamma corrected to generate a spatially adaptive co-occurrence weight map. An energy functional is constructed that includes an infrared image fidelity term, a visible light image fidelity term, and a total variation regularization term weighted by a spatially adaptive co-occurrence weight graph, to obtain a weighted total variation model; The weighted total variation model is iteratively solved using the alternating direction multiplier method to obtain the final fused image; Specifically, based on the quantized infrared image, a local gray-level co-occurrence matrix is ​​constructed for each pixel, and the co-occurrence probability is calculated based on the local gray-level co-occurrence matrix, including: For each pixel in the quantized infrared image, define a point centered on it. A local neighborhood window of a specific size; It is a positive integer greater than 1; Within each local neighborhood window, calculate Gray-level co-occurrence matrix of size , where matrix elements This indicates that the grayscale value within this window is... The pixel and grayscale values The number of times that pixels appear together according to a preset spatial offset; This is the preset quantization level for grayscale quantization of infrared images; gray-level co-occurrence matrix Normalization is performed to obtain the co-occurrence probability matrix corresponding to the local neighborhood window. , where matrix elements Indicates at pixel point Within a local neighborhood, grayscale pairs The probability of occurrence of adjacent targets in the same direction; The initial weight of each pixel is calculated based on the co-occurrence probability, specifically including: For the current pixel, form a gray-level pair between the gray-level value of the pixel and the gray-level values ​​of all other pixels in its local neighborhood window; In the co-occurrence probability matrix In the middle, find the co-occurrence probability value corresponding to each gray-level pair; The initial weight of a pixel is obtained by summing the co-occurrence probability values ​​of all grayscale pairs.

2. The infrared and visible light image fusion method based on the co-occurrence statistically induced weighted total variation model according to claim 1, characterized in that, Gray-level quantization of infrared images specifically includes: Quantize the original grayscale range of the infrared image to Each level, the quantized image any pixel pixel grayscale value Calculated using the following formula: in, This indicates the original infrared image at the pixel level. The pixel grayscale value at that location, This indicates the floor function.

3. The infrared and visible light image fusion method based on the co-occurrence statistically induced weighted total variation model according to claim 2, characterized in that, Quantitative Level It is 128.

4. The infrared and visible light image fusion method based on the co-occurrence statistically induced weighted total variation model according to claim 1, characterized in that, The range of the correction coefficients for Gamma correction is (0, 1).

5. The infrared and visible light image fusion method based on the co-occurrence statistically induced weighted total variation model according to claim 1, characterized in that, The expression for the energy functional is: in, For the desired fused image, The original infrared image, Visible light image, These are the weighting coefficients for the infrared and visible light fidelity terms, respectively. For regularization parameters, For spatial adaptive co-occurrence weight graph, For local neighborhood windows, To merge images gradient magnitude, This represents the Frobenius norm.

6. The infrared and visible light image fusion method based on the co-occurrence statistically induced weighted total variation model according to claim 5, characterized in that, The weighted total variation model is iteratively solved using the alternating direction multiplier method to obtain the final fused image, specifically including: Introducing auxiliary variables auxiliary variables As a constraint in the energy functional minimization problem; Construct an augmented Lagrangian function for the constrained energy functional minimization problem; Based on the augmented Lagrangian function, the solution of the minimization problem is decomposed into a problem concerning the fused image. Sub-problems and about auxiliary variables The two subproblems are solved iteratively and alternately until the convergence condition is met, and the final fused image is output. .

7. The infrared and visible light image fusion method based on the co-occurrence statistically induced weighted total variation model according to claim 6, characterized in that, Auxiliary variables The subproblem is solved by the weighted soft threshold shrinkage operator.

8. The infrared and visible light image fusion method based on the co-occurrence statistically induced weighted total variation model according to claim 7, characterized in that, The threshold of the weighted soft threshold shrinkage operator is ,in For spatial adaptive co-occurrence weight graph, To augment the penalty parameters of the Lagrange multiplier method.

Images