A digital camouflage pattern design method based on trend function constraint

Abstract

This invention belongs to the field of digital camouflage technology, specifically relating to a digital camouflage pattern design method based on trend function constraints. It includes: Step 1: Pre-segmenting the background image, then obtaining the step size, turning angle, and step size probability distribution P. l and the probability distribution of the turning angle P θ Step 2: Normalize the frequency and pixels of the background image, obtain the weights corresponding to the edges, construct a Laplacian sparse matrix, generate a probability matrix, match patch labels, calculate the density of texture patches, adaptively generate corresponding convolutional modules, and finally design destructive edges; Step 3: Perform matching and clustering of the main color to output the final camouflage pattern. This invention proposes a digital camouflage pattern design method based on trend function constraints. By defining a trend function to estimate the boundary probability between pixel nodes, digital camouflage patches are outlined, and finally, the background main color is filled to distort the target surface features, thereby realizing the digital camouflage pattern design.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of digital camouflage, and particularly relates to a digital camouflage pattern design method based on trend function constraint. BACKGROUND

[0002] In the face of high-tech imaging reconnaissance, battlefield detection means tend to be automated and refined, and digital camouflage is a common means of countering military reconnaissance. The degree of optical camouflage of military sites, mobile personnel on the battlefield and special mobile equipment in wartime depends largely on the advantages and disadvantages of camouflage pattern design.

[0003] In an optical image, a pixel is regarded as the smallest unit that cannot be divided in the entire image. Compared with the smooth edges of traditional camouflage, digital camouflage composed of mosaics of different shapes and sizes shows better camouflage effects in various scenes. Modern information-based warfare is dominated by information technology and mainly uses information-centric warfare. The most intuitive manifestation is the increasing value of time, which makes "decisive battle at first, destruction upon discovery". Such "instant kill" era brings unprecedented challenges to the battlefield survival capability of main battle equipment. Therefore, for different geographical environments, special camouflage uniforms, camouflage faces and camouflage nets need to be provided for combat personnel and equipment.

[0004] Digital camouflage design mainly includes two parts: background color tone extraction and camouflage patch design. The main color extraction of the background image generally uses K-means clustering method, color histogram method, etc. According to the similarity rule, the classification of image colors can obtain the main representative colors in the background image. Filling the designed digital patches can increase the similarity between the digital camouflage pattern and the original background and enhance the camouflage effect of the target.

[0005] The design of traditional camouflage often mainly depends on the experience of designers, and there is a certain difference between the design and the surrounding background of the camouflage target. In addition, the strong distinction between different colors weakens the camouflage effect and makes it difficult to cope with high-resolution reconnaissance means. Considering the fixed invariability of two-dimensional digital camouflage, the texture lines of visual illusion strips are introduced into the design of camouflage patches, and the camouflage patterns are generated by arranging and combining according to certain rules, which stimulates the visual perception deviation of human vision to a certain extent, increases the deceptive nature of the camouflage target at some observation angles, and weakens the similarity between the digital camouflage pattern itself and the background. SUMMARY

[0006] In view of the above existing problems, the application provides a digital camouflage pattern design method based on a trend function constraint, a boundary probability between pixel nodes is estimated by defining a trend function, a digital camouflage patch is outlined, and finally a background main color is called to distort target surface features, so that the digital camouflage pattern design is realized.

[0007] In order to achieve the above object, the technical scheme adopted by the application is as follows:

[0008] A digital camouflage pattern design method based on a trend function constraint comprises the following steps:

[0009] Step 1: pre-separating a background image, then obtaining a step length, a corner, and a step length probability distribution P l and a corner probability distribution P θ ;

[0010] Step 2: on the basis of step 1, performing frequency normalization and pixel normalization on the background image, obtaining an edge corresponding weight, then constructing a Laplace sparse matrix and generating a probability matrix, then matching patch labels, calculating a texture patch density, generating a corresponding convolution size according to the density corresponding to different positions in the background image, and designing a destructive edge;

[0011] Step 3: on the basis of the processing of step 2, using a K-means clustering method to match and cluster main colors of the background image, and outputting a final camouflage pattern.

[0012] Preferably, step 1 comprises:

[0013] Step 1.1: regarding each pixel in the background image as a node N x,y , wherein x and y respectively represent row and column coordinates of the background image, a connection line between nodes is defined as an edge E(N x,y , N x',y' ), the background image is divided by the connection of nodes and edges, and a digital camouflage patch is obtained;

[0014] Step 1.2: randomly selecting a pixel point of the background image as a moving point M, taking the current node as a vertex, moving the moving point M to the next node as a terminal point, setting a ray pointing to the horizontal right side of the image from the vertex as a starting edge, a connection line from the vertex to the terminal point as a terminal edge, and an included angle between the terminal edge and the starting edge clockwise rotating to the direction of the starting edge as a corner θ;

[0015] Step 1.3: The background image is segmented in step 1.1, and the segmented image is processed by a morphological algorithm to remove noise points. The obtained information of each segmented edge point is counted, and the 8-neighborhood direction of each pixel point is iterated. During iteration, if it is the same as the previous step, the step length l+1 is counted, and if it is different, the turning angle θ is calculated. After traversing the entire background image, the step length statistical value L={l1, l2, …l max} and the direction turning angle statistical value are obtained, and then the proportion of different step lengths l and angles θ is calculated. The frequency of occurrence of different step lengths and the frequency of occurrence of different angles are counted, and the step length probability distribution P l and the turning angle probability distribution P θ are obtained.

[0016]

[0017] Preferably, step 2 is specifically:

[0018] Step 2.1: According to the step length probability distribution P l and the turning angle probability distribution P θ obtained in step 1, the walking step length is randomly generated according to the probability and normalized to obtain the step length parameter S stepPara :

[0019] S stepPara =R Random (Φ(L),P l );

[0020] Where Φ(·) represents normalization processing, and R Random (Φ(L),P l ) represents generating a certain data in the corresponding set Φ(L) according to the probability value in P l .

[0021] Step 2.2: The coordinates of all pixel points in the background image are converted into line indexes, and the edge index E(x, y) is set according to the coordinate order. According to the edge index assumption in the background image, the pixel difference of each potential edge is calculated and normalized, and the pixel parameter V valPara is:

[0022]

[0023] In the formula, I val is the pixel value of the current position, E(x', y') and E(x, y) represent the edge indexes of the coordinate positions (x' ,y ') and (x, y), respectively, and (x', y') represents the horizontal neighborhood coordinates or vertical neighborhood coordinates when the current coordinate is (x, y);

[0024] Step 2.3: On the basis of steps 2.1 and 2.2, the Gaussian weight information W corresponding to each edge index is calculated:

[0025] W = exp(-(a · V valPara + b · S stepPara ))+ C

[0026] where a and b are the coefficients of V valPara and S stepPara parameters respectively, C is the minimum selectable constant, Sparse(·) represents generating a sparse matrix.

[0027] Step 2.4: According to the edge index and the corresponding Gaussian weight information of each item in step 2.3, and the number of edge indexes, a Laplace sparse matrix Γ is constructed:

[0028] Γ = diag(∑S)-S

[0029] where the matrix S is composed of edge indexes and corresponding edge weight information, and the row and column are equivalent to the sparse square matrix of the total number of pixel points, diag(·) represents constructing a diagonal matrix;

[0030] At this time, the matrix Z used to estimate the edge of the texture patch is represented as,

[0031]

[0032] where Γ core is a sparse matrix containing the core pixel points of k camouflage patches, is a non-core sparse matrix, I k is a k-order identity matrix;

[0033] Step 2.5: Set the Dirichlet problem, solve the combined Dirichlet problem to solve the Laplace sparse matrix, and obtain the probability matrix. According to the returned maximum value and index information of the row and column, match the texture patch label;

[0034] Step 2.6: Calculate the size D of the patch, that is, the number of image pixels occupied by each patch. According to the distance of each pixel position from the patch boundary, the density of each pixel point is calculated, and all values are normalized. Then according to the density size, the convolution size a of each position participating in the replacement is designed, which is obtained by

[0035] Preferably, the Dirichlet problem of step 2.5 and the step of obtaining the probability matrix by solving the Dirichlet problem are specifically:

[0036] Step 2.5.1: According to the number of patches obtained by pre-segmenting the background image, the entire background image is equally divided into k, and the center position of each equal division area is defined as the core pixel point position of the camouflage patch.

[0037] ​Step 2.5.2: Define the k-order unit matrix, that is, initialize the boundary matrix;

[0038] Step 2.5.3: Based on the sparse matrix of non-core pixel points That is, the information of the core pixel point position corresponding to the camouflage patch in the Laplace matrix is set to empty, the Laplace sparse matrix Γ of the core pixel point core , and the boundary matrix I is initialized k The matrix Z for estimating the edge of the texture patch is obtained, that is, the probability matrix, which has a size of m x k, where m is the number of all pixel points of the background image.

[0039] Preferably, the step 2.5 of matching the texture patch label according to the returned row and column maximum value and index information is specifically:

[0040] According to the maximum value of each row in the probability matrix, that is, the probability, and the column number, that is, the core seed number, that is, the label number of the subsequent camouflage patch, the index information vector of the obtained n numbers is reconstructed in size according to the original background image size, and the final matching texture patch label is obtained.

[0041] Compared with the prior art, the beneficial effects of the present application are:

[0042] (1) The effectiveness of digital camouflage in achieving camouflage lies in its chaotic texture and patches. After coloring the digital patches generated based on random walk, good destructiveness can be achieved in the entire scene, but the original texture trend of the background pattern is ignored. In this paper, biased random walk is adopted, that is, while preserving the information of pixel value, texture corner direction and walk step length, the random walk with "trend" is completed according to a certain probability, realizing the certainty and randomness of digital camouflage pattern design for specific background;

[0043] (2) Inspired by animals avoiding predators in nature, animals gradually evolve patterns similar to the background to adapt to the environment and improve their survival ability. In this paper, a self-adaptive convolution template is used to destroy the edges of the generated texture patches based on dense mapping, so as to achieve the purpose of upgrading camouflage. BRIEF DESCRIPTION OF DRAWINGS

[0044] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application.

[0045] In the drawings:

[0046] Figure 1 is the overall design flowchart of the present application;

[0047] Figure 2 is the method flowchart of the present application;

[0048] Figure 3 Define a schematic diagram for the direction angle;

[0049] Figure 4 An example diagram of the present application. DETAILED DESCRIPTION

[0050] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, in which it is understood that the preferred embodiments described herein are merely intended to illustrate and explain the present application, and are not intended to limit the present application.

[0051] Embodiments:

[0052] Referring to the accompanying drawings, a digital camouflage pattern design method based on trend function constraint includes: Figures 1-4 As shown in the drawings, a digital camouflage pattern design method based on trend function constraint includes:

[0053] Step 1: Pre-segment the background image, and then obtain the step length, the turning angle, and the step length probability distribution P l and the turning angle probability distribution P θ ; including:

[0054] Step 1.1: Set each pixel in the background image as a node N x,y , where x and y represent the row and column coordinates of the background image, respectively, the connection between nodes is defined as an edge E(N x,y , N x',y' ), and the background image is segmented by connecting the nodes and edges to obtain a digital camouflage patch; a good digital patch should have both the certainty of the background texture and the randomness of the target survival scene. Therefore, the selection of nodes and the direction of edges are important in the generation process of the camouflage patch.

[0055] Step 1.2: Randomly select a pixel point of the background image as a moving point M, the running track of the moving point M in the image is piecewise linear, according to the general assumption of motion type, this "linearity" depends on the motion time and position of each point, which is equivalent to the node coordinates N x,y =(x,y) and the edge length E(N x,y , N x',y'). And "segment" is reflected in the angle between the edge and the edge, with the current node as the vertex, the moving point M moves to the next node as the terminal point, and the ray from the vertex pointing to the horizontal right side of the image is set as the initial edge, the connecting line from the vertex to the terminal point is the terminal edge, and the angle between the terminal edge and the initial edge clockwise to the direction of the initial edge is recorded as the turning angle θ. Thus, the texture curve drawn in the image is the digital camouflage patch. Different from the digital camouflage patch obtained in step 1.1, the step 1.2 is to generate a "regular" "irregular" camouflage patch texture according to the idea of random walk based on the statistical analysis of the result obtained in 1.1. The digital camouflage patch obtained in step 1.1 is pre-segmented, and simply depends on the feature information of the original background image.

[0056] Step 1.3: The overall idea of random walk is to stop at (x l ,y l ) with a given probability P(x,y,θ,l) (l is the movement distance) at (x x,y,θ ,y ).

[0057]

[0058] In the formula, ζ x,y,θ (s) represents the vector of the movement unit starting from (x,y) in the distance interval [x+s x ,y+s y ], and ζ x,y,θ (s) = s·∠θ. Based on the biased characteristic of the chemotactic factor ρ, a simple function correlation can be derived:

[0059] P(x,y) = P(ρ(x,y)),

[0060] This means that each movement unit can only be affected by the chemotactic factor at a specific location. Based on this, a simplest dependence model is assumed:

[0061]

[0062] Where c is a constant.

[0063] Based on the above dependence model, various mathematical model mechanisms can be established to enrich the chemotactic function ρ(x,y), thereby realizing the control of the biased random walk of the seeds in the movement unit.

[0064] Considering that the original background texture information needs to be preserved for the camouflage patch, the constraint function first learns the prior information of the background texture. In step 1.1, the background image is segmented, and the morphological algorithm is used to remove the noise points of the segmented image. The information of each segmented edge point is counted, and the 8-neighborhood direction of each pixel point is iterated. During iteration, if it is the same as the previous step, the step length l+1 is counted, and if it is different, the corner angle θ is calculated. After traversing the entire background image, the step length statistical value L={l1, l2, …l max} and the direction corner statistical value are obtained, and then the proportion of different step lengths l and angles θ is calculated. The frequency of occurrence of different step lengths and the frequency of occurrence of different angles are counted, and the step probability distribution P l and the corner probability distribution P θ are obtained:

[0065]

[0066] The direction corner is defined as shown in Figure 2 . As can be seen from the figure, the pixel points in the same straight line direction have a corresponding relationship in position, and repeated calculation will occur when the direction corner is counted. Therefore, when setting the random walk precondition, the 8 direction corners are converted into 4 direction corners, and batch iteration is considered to ensure that there is no missing or overlap in the random walk adjacent step length.

[0067] Step 2: On the basis of step 1, the frequency normalization and pixel reduction of the background image are performed to obtain the edge corresponding weight, and then the Laplace sparse matrix is constructed, and the probability matrix is generated. Then the patch label is matched, the texture patch density is calculated, the corresponding convolution module is adaptively generated, and finally the destructive edge is designed; Specifically:

[0068] Step 2.1: According to the step length probability distribution P l and the corner probability distribution P θ obtained in step 1, the walk step length is randomly generated and normalized to obtain the step length parameter S stepPara :

[0069] S stepPara =R Random (Φ(L),P l );

[0070] Where Φ(·) represents the normalization processing, R Random (A·B) represents generating a certain data in set A according to the probability value in set B, and A and B are in a corresponding relationship;

[0071] Step 2.2: The core of realizing random walk is to get the weight of each texture edge (straight edge), in order to comprehensively describe the background information, considering the pixel value of the current position and the geometric distance (step size and direction). All pixel coordinates in the background image are converted into line index, and edge index E(x, y) is set according to the coordinate order. According to the edge index assumption in the background image, the pixel difference of each potential edge is calculated and normalized, and the pixel parameter V valPara is:

[0072]

[0073] In the formula, I val is the pixel value of the current position, I val is the pixel value of the current position, E(x', y') and E(x, y) represent the edge index of the coordinate position (x', y') and (x, y) respectively, and (x', y') represents the horizontal neighborhood coordinate or the vertical neighborhood coordinate when the current coordinate is (x, y);

[0074] Step 2.3: On the basis of step 2.1 and step 2.2, the Gaussian weight information W corresponding to each edge index is calculated:

[0075] W = exp(-(α·V valPara + β·S stepPara ))+ C

[0076] Wherein, α and β are the coefficients of V valPara and S stepPara parameters, C is the minimum selectable constant, which is used to ensure the numerical stability, and Sparse(·) represents the generation of sparse matrix.

[0077] Step 2.4: According to the edge index and the corresponding Gaussian weight information of step 2.3, and the number of edge index, the Laplace sparse matrix Γ is constructed:

[0078] Γ = diag(∑S)-S;

[0079] Wherein, the matrix S is composed of edge index and corresponding edge weight information, and the row and column are equivalent to the sparse square matrix of the total number of pixel points, and diag(·) represents the construction of diagonal matrix;

[0080] At this time, the matrix Z used to estimate the texture patch edge is represented as,

[0081]

[0082] Wherein, Γ core is the sparse matrix of the core pixel points containing k camouflage patches, is the non-core sparse matrix, I k is the k order unit matrix.

[0083] Step 2.5: Set the Dirichlet problem, solve the Laplace sparse matrix by solving the combined Dirichlet problem, and obtain the probability matrix. According to the returned row and column maximum value and index information, match the texture patch label.

[0084] The steps of the Dirichlet problem and the probability matrix obtained by solving the Dirichlet problem are as follows:

[0085] Step 2.5.1: According to the number of patches obtained by pre-segmenting the background image, divide the entire background image into k equal parts, and define the center position of each equal part as the core pixel position of the camouflage patch.

[0086] Step 2.5.2: Define the k-order unit matrix, i.e. initialize the boundary matrix.

[0087] Step 2.5.3: Based on the sparse matrix of non-core pixel points Set the information of the core pixel position of the camouflage patch in the Laplace matrix to empty, the Laplace sparse matrix Γ core of the core pixel point, and initialize the boundary matrix I k Obtain the matrix Z used to estimate the edge of the texture patch, i.e. the probability matrix, which has a size of m x k, where m is the number of all pixel points in the background image.

[0088] According to the returned row and column maximum value and index information, the matching texture patch label is as follows:

[0089] According to the probability matrix, obtain the maximum value of each row, i.e. the probability, and the column number, i.e. the core seed number, which is the label number of the subsequent camouflage patch. According to the size of the original background image, the index information vector of the obtained n numbers is reconstructed in size to obtain the final matching texture patch label.

[0090] Step 2.6: Calculate the texture patch density, and generate the corresponding convolution size according to the density corresponding to different positions in the background image, and design the destructive edge.

[0091] Specifically: count the size of the patch D (i.e. the number of image pixels occupied by each patch), calculate the density of each pixel point according to the distance of each pixel position from the patch boundary, and normalize all values. Then, according to the density size, design the convolution size a of each position participating in replacement, which can be calculated as , which actually means the average value of all patch sizes in the image, and the more dense the patch, the smaller the convolution size in the design process of the destructive edge. In addition, not every point in the convolution size participates in pixel replacement, and the density also represents the probability of each pixel point participating in pixel replacement.

[0092] Step 3: Based on the processing in Step 2, perform primary color matching and clustering to output the final camouflage pattern. Specifically:

[0093] First, select K initial cluster centers, i.e., K initial colors. Each color includes pixel values ​​for three channels: R (red), G (green), and B (blue). Calculate the Euclidean distance between each pixel and the K initial cluster centers, and group pixels according to the closest distance. After grouping, use the mean value of all points in each class as the new center of that class. Iterate and update the cluster centers by calculating the distance of each pixel to the center. Once the results stabilize, output the K primary colors for each cluster. Finally, based on the proportion of each primary color in the background image, randomly fill the marked texture patches obtained in step 2 to generate the final digital camouflage pattern.

[0094] Inspired by chameleons, squid, and other "masters of camouflage" in the animal kingdom, and considering the destructive effect of gradient edges on texture, this paper proposes adding irregularly scattered dots to the edges of patches. Since the labels on the background image region markers reflect the distribution of patches in the background, the density is calculated based on the patch size information.

[0095] To make the designed camouflage pattern destructive to targets, its randomness is increased in two ways. First, the density map of texture patches is converted into a probability map. That is, areas with higher brightness in the density map indicate smaller patch sizes and higher texture density, while areas with lower brightness in the density map have relatively larger patch sizes and sparser texture. Therefore, the size of the scattered template in the area is adaptively designed according to this rule. Second, the proportion of the extracted background main color in the original background is calculated, and random numbers are generated according to the proportion. This ensures that the main color is filled in to match the background color distribution, thereby increasing the camouflage effect of the digital pattern.

[0096] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A digital camouflage pattern design method based on trend function constraints, characterized in that: include: Step 1: Pre-segment the background image, then obtain the step size, rotation angle, and step size probability distribution. and the probability distribution of turning angles ; Step 2: Based on Step 1, frequency and pixel normalization of the background image are performed to obtain the edge weights. Then, a Laplacian sparse matrix is ​​constructed, and a probability matrix is ​​generated. Next, patch labels are matched, and the density of texture patches is calculated. Based on the density corresponding to different positions in the background image, the corresponding convolution size is generated, and destructive edges are designed. Specifically, Step 2 involves: Step 2.1: Based on the step size probability distribution obtained in Step 1 and the probability distribution of turning angles The walk step size is randomly generated according to probability and then normalized to obtain the step size parameter. for: ； in, This indicates normalization processing. Indicates by Generate a corresponding set from the probability values ​​in the set. A certain data in; Step 2.2: Convert the coordinates of all pixels in the background image to line indices, and set the edge indices in coordinate order. Based on the edge index assumption in the background image, the pixel difference of each potential edge is calculated and normalized, and the pixel parameters are... for: ； In the formula, This is the pixel value at the current position. and Representing coordinate positions respectively and edge index, current coordinate is hour, Represents the horizontal or vertical neighborhood coordinates; Step 2.3: Based on Steps 2.1 and 2.2, calculate the Gaussian weight information W corresponding to each edge index: ； in, and They are respectively and The coefficients of the parameters, where C is the smallest possible constant; Step 2.4: Based on the edge indices and corresponding Gaussian weights from Step 2.3, and the number of edge indices, construct the Laplacian sparse matrix. : ； Here, matrix S consists of edge indices and corresponding edge weights, and its rows and columns are equivalent to a sparse square matrix representing the total number of pixels. This indicates the construction of a diagonal matrix; At this point, the matrix used to estimate the edges of texture patches Represented as, ； in, Given a sparse matrix containing the core pixels of k camouflage patches. It is a non-core sparse matrix. It is a k-order identity matrix; Step 2.5: Set up the Dirichlet problem, solve the combined Dirichlet problem to solve the Laplace sparse matrix, obtain the probability matrix, and match the texture patch markers based on the returned row and column maximum values ​​and index information; Step 2.6: Calculate the size D of each patch, i.e., the number of image pixels occupied by each patch. Calculate the density of each pixel based on its distance from the patch boundary, and normalize all values. Then, based on the density, design the convolution size 'a' for each location participating in the permutation. Calculated; Step 3: Based on the processing in Step 2, the K-means clustering method is used to match and cluster the main colors of the background image to output the final camouflage pattern.

2. The digital camouflage pattern design method based on trend function constraints according to claim 1, characterized in that: Step 1 includes: Step 1.1: Let each pixel in the background image be a node. Where x and y represent the row and column coordinates of the background image, respectively, and the lines connecting the nodes are defined as edges. The background image is segmented by connecting nodes and edges to obtain digital camouflage patches; Step 1.2: Randomly select a pixel in the background image and treat it as a moving point M. With the current node as the vertex and the next node as the endpoint, define the ray from the vertex to the horizontal right side of the image as the starting edge, and the line connecting the vertex and the endpoint as the ending edge. The angle between the ending edge and the starting edge when rotated clockwise is recorded as the rotation angle. ; Step 1.3: The background image is segmented in Step 1.1, and a morphological algorithm is used to remove noise from the segmented image. The information of each segmentation edge point is statistically analyzed. The 8-neighborhood direction of each pixel is iterated. During iteration, if it is the same as in the previous step, the step size l+1 is calculated; if it is different, the rotation angle θ is calculated. After traversing the entire background image, the step size statistics are obtained. The step length probability distribution is obtained by statistically analyzing the direction and angle, calculating the ratio of different step lengths l to angles θ, and statistically analyzing the frequency of different step lengths and different angles. and the probability distribution of turning angles : 。 3. The digital camouflage pattern design method based on trend function constraints according to claim 2, characterized in that: The specific steps for solving the Dirichlet problem in step 2.5 and obtaining the probability matrix are as follows: Step 2.5.1: Based on the number of patches obtained from the pre-segmentation of the background image, divide the entire background image into k equal parts, and define the center position of each equally divided region as the core pixel position of the camouflage patch; Step 2.5.2: Define the k-order identity matrix, i.e., initialize the boundary matrix; Step 2.5.3: Sparse matrix based on non-core pixels That is, the information of the core pixel position of the corresponding camouflage patch in the Laplacian matrix is ​​set to empty, and the Laplacian sparse matrix of the core pixel is used. and initializing the boundary matrix The matrix Z used to estimate the edges of texture patches is obtained, which is the probability matrix with a size of m×k, where m is the total number of pixels in the background image.

4. The digital camouflage pattern design method based on trend function constraints according to claim 3, characterized in that: Step 2.5, which involves matching texture patch markers based on the returned row and column maximum values ​​and index information, specifically involves: The maximum value corresponding to each row is obtained from the probability matrix, which is the probability, and the column number, which is the core seed number and the label number of the subsequent camouflage patch. The vector of index information of the n numbers is reconstructed according to the size of the original background image to obtain the final matching texture patch label.

Images