Digital camouflage generation method based on noise

Abstract

The invention provides a noise-based digital camouflage generation method, which comprises the following steps of: obtaining a scene color set by virtue of a color extraction algorithm through five environment images from different directions of a camouflage target, re-customizing and optimizing a genetic algorithm bottom layer operator aiming at a noise generation and superposition process parameter tuning problem, and obtaining a scene color set; on the basis of taking a newly-defined camouflage comprehensive evaluation factor as fitness, key parameters of a Perlin noise superposition texture generation process are adjusted and optimized, meanwhile, parameter adjustment and optimization ranges are limited for different types of camouflage objects, the optimal camouflage texture of a scene is generated after adjustment and optimization, then digital camouflage is generated in a digitized mode, and the optimal camouflage texture of the scene is obtained. And the final camouflage is obtained through further adjustment according to national standard specifications. The camouflage generation efficiency is greatly improved, and the self-adaptive camouflage effect of various camouflage objects in multiple scenes is improved.

Description

Technical Field

[0001] The invention belongs to digital camouflage generation technology, and in particular relates to a noise-based digital camouflage generation method. Background Art

[0002] Camouflage is crucial in modern warfare and military activities. Its design purpose is to minimize the enemy's ability to detect and identify our armored vehicles in different environments.

[0003] The overall camouflage design is divided into two parts: background color extraction and pattern texture design. Traditional camouflage design relies on manual color selection based on the background, and then draws the camouflage pattern in the scene based on experience. The development of computer vision and image processing technology and the improvement of computing power have made it possible to automatically generate camouflage textures that are highly adaptable to specific environments.

[0004] There are many generation methods at this stage, mainly focusing on camouflage generation based on deep learning style transfer, specific scene camouflage based on adversarial neural networks, image recognition and environment clustering, etc. However, these existing studies and related methods still have certain defects, such as: low texture generation efficiency, too long early optimization time, and extremely limited by computing resources, poor adaptability to multiple scene environments, and the verification and evaluation standards used also have certain defects, resulting in the actual camouflage effect of the final camouflage not meeting expectations. The above are all problems that need to be solved in the field of camouflage generation, and are also the starting point for the method of the present invention. Summary of the invention

[0005] The purpose of the present invention is to provide a noise-based digital camouflage generation method, through a customized improved genetic algorithm intervention optimization method for multi-Perlin noise texture generation, the underlying operators of the genetic algorithm, especially the probability calculation of each operator, are redesigned and optimized, and a new multi-dimensional texture evaluation factor is proposed as fitness, and the interpolation algorithm in the texture generation framework is improved, so that the tuning efficiency of texture generation parameters is improved, and the texture generation effect after tuning is also improved, so that better camouflage textures can be stably output, and the background environment can be better integrated, and the camouflage effect can be more significantly improved compared with other algorithms.

[0006] The embodiments of the present invention can be implemented as follows:

[0007] The present invention provides a noise-based digital camouflage generation method, which is used to generate a camouflage digital camouflage suitable for a target scene according to a target scene. The generation method comprises:

[0008] In non-rainy or snowy weather with sufficient natural light, capture environmental images in the target scene from five directions: front, back, left, right, and bottom of the shooting point;

[0009] Extracting the final main color set members in the target scene from the five environment images for color distribution planning of the background, main spots, modified spots, etc. in the camouflage texture;

[0010] Select multiple Perlin noise superpositions to generate camouflage textures and optimize the internal algorithm, further screen the parameters involved in the noise texture generation algorithm process, and obtain the key process parameters to be tuned;

[0011] According to the parameter tuning problem in the noise texture generation process, the underlying operator of the genetic algorithm is reconstructed, and a more comprehensive camouflage evaluation factor is designed as the fitness of the genetic algorithm for the tuning process of the key parameters of the process, wherein the optimization direction is specifically restricted for different types of camouflaged objects to adapt to the camouflage requirements of the camouflaged objects in the current target scene background, and to integrate into the environment background as much as possible;

[0012] The tuned parameters are locally stored together with the environment image, and each time a new image is acquired in the target scene, the image is evaluated and compared with the locally stored image to determine whether the parameter update or the parameter is directly used to generate a deformed camouflage, and then a digital camouflage texture module is digitally generated;

[0013] According to the national camouflage standards such as segmentation attributes and layering attributes, the layout of the digital camouflage texture module, the relationship between the main spots, the spot modification effect, etc. are adjusted.

[0014] Further, in an optional embodiment, the specific process of extracting colors to obtain the final primary color set and planning color distribution is:

[0015] Using a clustering algorithm on the color data in the five environment images respectively, clustering pixels with similar colors into the same category, and outputting the representative color of the category, which is recorded as one color;

[0016] Similarly, the first 4 colors of each environment image are obtained and stored in color set C. Five color sets C are obtained from five images. i ,i∈{1,2,3,4,5},c mn Represents the color set C obtained by the mth image m The color with the nth largest proportion;

[0017] If for c mn ∈C i , then c mn Put it into the final color set G, and finally get the final color set G with x elements. If x<4, take

[0018]

[0019] The color g n The final color set G is also stored, so that the final color set G contains 4 color elements as the final color set of the target scene;

[0020] The first color in the final color set G is used as the main background color of the camouflage texture, the second and third colors are used as optional colors for generating the main spots, and the fourth color is reserved as a color for adding additional modification spots.

[0021] Further, in an optional embodiment, the specific steps of optimizing and adjusting the internal algorithm to generate camouflage texture by superimposing multiple Perlin noises and screening key parameters of the process are as follows:

[0022] After the early steps of noise calculation such as grid network definition, gradient vector allocation, distance vector calculation, and dot product calculation, when the dot product values ​​of adjacent grid points are combined through an interpolation function, the present invention introduces a texture adjustment factor α in conventional bilinear interpolation. The noise calculation of the conventional bilinear interpolation is:

[0023] N(x,y)=lerp(u,lerp(v,dot i,j ,dot i+1,j ),lerp(v,dot i,j+1 ,dot i+1,j+1 ))

[0024] The improved interpolation process after introducing the texture adjustment factor is:

[0025]

[0026] Where lerp(t,a,b) represents the linear interpolation function, dot i,j 、dot i+1,j 、dot i,j+1 、dot i+1,j+1 Respectively represent the dot product between the gradient vector of the four vertices, namely the bottom left vertex, the bottom right vertex, the top left vertex, and the top right vertex, and the distance vector from the vertex to the interpolation point (x, y). They represent the noise values ​​obtained after adjusting the dot product values ​​of each vertex. u and v are the smoothing function values ​​in the x and y directions based on the input point coordinates. The smoothing function uses an improved 5th-order Hermite smoothing function to ensure better interpolation smoothness.

[0027] By adjusting the value of α, you can create richer and more delicate effects in different texture areas, thereby better simulating the variability and complexity of camouflage textures in nature. α will also be an important parameter to be tuned.

[0028] Then, multiple layers of Perlin noise with different frequencies and amplitudes are superimposed to generate complex and diverse textures. After the noise superposition is completed, color mapping is performed according to the final color distribution plan of claim 2 to generate the final camouflage texture.

[0029] The final parameters to be tuned after screening are specifically: the amplitude attenuation rate between different frequency layers in the noise texture, i.e., persistence, the texture adjustment factor α in the improved interpolation algorithm, the size of the feature in the noise texture, i.e., the noise texture detail ratio Scale, frequency Frequency, amplitude Amplitude, noise layer coefficient Octaves, lacunarity, offset offset, noise gain gain, threshold Threshold for binarization of noise values, Maximum value for limiting the maximum range of noise values, and Minimum value for limiting the minimum range of noise values.

[0030] Furthermore, in an optional embodiment, the improved genetic algorithm applied to this problem is specifically as follows:

[0031] The fitness of the improved genetic algorithm is a new camouflage evaluation factor, and the calculation method is:

[0032] Calculate the histogram of the final camouflage texture image and the environment image, and use the Bhattacharyya coefficient to compare the similarity between them. The larger the coefficient, the higher the similarity between the two images in color;

[0033] GLCM is used to evaluate the texture similarity between the final camouflage texture image and the environment image. GLCM is a statistical method for analyzing image texture features. It describes the texture features of an image by calculating the grayscale difference of pixels in different directions. The GLCM comparison function is used to calculate the similarity between two sets of GLCM. The larger the value, the higher the texture similarity between the two images.

[0034] Use Canny edge detection to evaluate the confusion effect of the final camouflage texture image and the environment image. Canny edge detection is a commonly used edge detection algorithm that can detect obvious edges in an image. Perform Canny edge detection on the two images respectively, and then compare their edge information. The smaller this value is, the better the confusion effect of the camouflage texture image and the environment background image on the edge is;

[0035] The scores of the three dimensions are averaged to obtain a comprehensive score, and the range is controlled to be between 0 and 1, which is used as a camouflage evaluation factor, that is, the fitness of the improved genetic algorithm.

[0036] The selection operator of the improved genetic algorithm is improved based on the ordinary roulette selection algorithm, and the new evaluation factor is used as the fitness. Through a large number of experiments, it is found that the texture evaluation factor with excellent effect is usually above 0.8, so the individuals with fitness higher than 0.8 are marked as excellent individuals, and the probability of the excellent individuals is increased to two times, making it easier to be selected and pass on the excellent features, that is, the probability P of each chromosome being selected is calculated each time i , there are M chromosomes in total, f i is the fitness of the chromosome, P i It can be expressed as

[0037]

[0038] The crossover method selected by the crossover operator of the improved genetic algorithm is single-point crossover, and the mutation method selected by the mutation operator is single-point mutation. Both the crossover probability and the mutation probability are dynamically adjusted according to the population state:

[0039] Crossover probability:

[0040] Where P c is the crossover probability, P c_max is the maximum crossover probability in the current population, f′ is the larger fitness value of the two parent individuals in the crossover operation, and f max is the maximum fitness value in the entire population, f min is the minimum fitness value. In this application, the fitness function is accuracy, so f∈[0,1],f max -f min ∈[0,1]

[0041] Mutation probability:

[0042] Where P m is the mutation probability, P m_max is the current maximum mutation probability, f is the fitness value of the individual, and the rest is the same as the crossover probability part above.

[0043] Furthermore, in an optional embodiment, the optimization range restrictions for different types of camouflage objects are specifically as follows:

[0044] For the uniforms, helmets, carrying gear, weapons and other personal equipment and uniform camouflage of relevant personnel, the range restrictions of the parameters such as Amplitude, Scale, α, etc. in claim 3 are added, specifically: Amplitude is between 0.5 and 2.0, Scale is between 10 and 100, and α is between 1 and 5;

[0045] For the camouflage of medium and long-range equipment such as armored vehicles, non-armored support vehicles, artillery systems, surface ships, fixed-wing aircraft, etc., add the range limits of the parameters such as Scale, Octaves, and Frequency as stated in claim 3, specifically: Scale is between 50 and 500, Octaves is between 4 and 8, and Frequency is between 0.01 and 0.1;

[0046] For the camouflage of fixed military facilities such as command posts, fortifications, logistics facilities, bridges and important transportation nodes, the range limits of the parameters such as Scale, Octaves, α, etc. in claim 3 are added, specifically: Scale is between 100 and 1000, Octaves is between 6 and 10, and α is between 2 and 6;

[0047] For submarine and other underwater equipment camouflage, add the range limits of parameters such as Scale, Octaves, and Persistence in claim 3, specifically: Scale is between 20 and 200, Octaves is between 5 and 9, and Persistence is between 0.4 and 0.7;

[0048] For tactical deception camouflage such as decoys and false targets, the range limits of parameters such as Octaves and α mentioned in claim 3 are added, specifically: Octaves is between 3 and 7, and α is between 3 and 8;

[0049] The above ranges are all closed ranges.

[0050] Furthermore, in an optional embodiment, the parameter update and specific application in an actual scenario include the following specific steps:

[0051] Input a new environment image collected from the target scene, use the first two evaluation dimensions described in claim 6 to perform color similarity evaluation and texture matching evaluation with the stored image of the scene, take the average of the two, and the final result ranges from 0 to 1; if there is no stored image, that is, it is the first time to run under the target scene, the evaluation result returned is 0;

[0052] When the evaluation result is less than 0.5, re-tuning is performed using the improved genetic algorithm of claim 7 to obtain a new parameter set, the new parameter set is used to overwrite and update the parameter set of the target scene, and the newly provided image of the current environment is overwritten to update the locally stored image of the target scene;

[0053] After the evaluation is completed, the parameter set is used to generate noise texture according to the specific steps in claim 4 to generate a final camouflage texture, that is, a deformed camouflage texture, and then a grid of appropriate size is constructed based on the deformed camouflage. When the pixels of the main pattern color in a certain grid are more than 50%, the grid is set to the main pattern color, and finally a digital camouflage texture is obtained;

[0054] Furthermore, in an optional embodiment, the digital camouflage module is adjusted according to the relevant camouflage standards, and the specific contents are as follows:

[0055] According to the segmentation requirements, if the distance between a main spot on a certain face of the camouflaged object and the edge is less than 10% of the width of the face in the direction perpendicular to the edge, the whole pattern is extended in the direction perpendicular to the edge and rotated clockwise until the overall direction of the spot pattern forms an angle of 30 degrees with the edge;

[0056] According to the multi-level requirements, for the main spot on a certain surface of the camouflaged object, calculate its maximum circumscribed rectangular area. If it is larger than 15% of the area of ​​the surface where the main spot is located, add peripheral modification spots of the same color to the outside and add inner modification spots of different colors to the inside; for the concave part of the camouflaged object, if the area exceeds 20% of the area of ​​the surface where the concave part is located, add the main spot to the concave part and use the brighter color member of the final color set described in claim 2; for the convex part of the camouflaged object, if the area exceeds 20% of the area of ​​the surface where the concave part is located, add the main spot to the convex part and use the darker color member of the final color set described in claim 2;

[0057] For an independent main spot on the camouflaged object, if there is a continuous rectangular area in the horizontal, vertical and oblique directions of the maximum circumscribed rectangle of the main spot without any main spots and modified spots, and the area of ​​the blank rectangular area exceeds 80% of the area of ​​the maximum circumscribed rectangle of the main spot, then other main spots and corresponding modified spots in a staggered, juxtaposed or other positional relationship are added to the blank rectangular area, that is, next to the original independent main spot.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] The invention selects a genetic algorithm with a relatively small number of parameter requirements, and customizes and improves the genetic algorithm based on three underlying operators of the genetic algorithm. When tuning parameters for texture generation, the invention quickly selects parameters with larger tuning weights from a small number of parameters such as amplitude, scale, octaves, persistence, frequency, an interpolation texture adjustment factor α, etc., and the gradient descent is more likely to approach the global optimal noise generation parameter set, which greatly accelerates the convergence speed, and improves the overall final tuning generation effect and parameter tuning efficiency. Through comparative experiments on various camouflage texture generation methods, the invention finds that the noise-generated camouflage texture is not only suitable for a high-complexity environment, but also has a better generation effect for a fine-grained environment than other methods, and has a better comprehensive camouflage effect. Therefore, the noise-generated texture is specially optimized for the purpose of camouflage, and a new interpolation scheme is especially improved on the basis of bilinear interpolation. After the texture adjustment factor α is newly introduced, the texture detail variation space is increased compared with the ordinary linear interpolation, so that the texture is more delicate and changeable, and is more suitable for the generation of camouflage texture in a fine-grained environment. Under the condition of reasonable parameter adjustment, it can be well adapted to the open natural environment with strong regularity. Therefore, bilinear interpolation is used as the interpolation function to innovate and improve the ordinary Perlin noise generated texture and the superimposed noise generated texture, forming an improved multi-noise superimposed texture generation that is more suitable for camouflage texture; the present invention proposes a new texture comprehensive evaluation factor as the fitness of the genetic algorithm, that is, the optimization index, to solve the defect that most existing evaluation methods only pay more attention to one aspect of color matching, texture similarity and confusion degree. How to accurately evaluate the effectiveness of generating camouflage is crucial to both process optimization and the effect of the final generated texture. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0061] Figure 1 A schematic diagram of a flow chart of a noise-based digital camouflage generation method provided by an embodiment of the present invention;

[0062] Figure 2 A schematic diagram of a color extraction process provided by an embodiment of the present invention;

[0063] Figure 3 Schematic diagram of the process of generating texture from noise and optimizing key parameters using an improved genetic algorithm provided by an embodiment of the present invention DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. However, the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0065] In one embodiment, Figure 1 As shown, Figure 1 1 is a flow chart of a method for generating digital camouflage based on noise provided by an embodiment of the present invention, comprising the following steps:

[0066] S101. In non-rainy or snowy weather with sufficient natural light, capture environmental images in the target scene from five directions, namely, front, back, left, right, and bottom, of a shooting point.

[0067] Common classic camouflage target scenes include deserts, woodlands, snowfields, grasslands, Gobi, etc. Try to choose a location with good visibility and distinct surrounding environment characteristics on a sunny day with plenty of light. Use conventional shooting equipment such as cameras and mobile phones to shoot in five directions, including forward, backward, left, right, and down, at a height of more than one meter from the ground to obtain the required environmental image.

[0068] S102: extracting a final main color set in the target scene from the five environment images, and using it for color distribution planning of the background, main spots, modified spots, etc. in the camouflage texture.

[0069] First, cluster the colors in each environment image to obtain several colors and obtain 5 color sets. Then find the colors that appear in all 5 color sets. If there are any, add them to the final color set. After this step, if the final color set does not reach 4 colors, use the high-frequency colors of the 5 color sets to find the average and supplement them. The final 4 colors are further planned for their specific uses in camouflage.

[0070] S103, selecting multiple Perlin noises to superimpose and generate camouflage textures and optimizing and adjusting the internal algorithm, further screening the parameters involved in the noise texture generation algorithm process, and obtaining key process parameters to be optimized.

[0071] First, the conventional noise calculation steps are performed: grid network definition, gradient vector allocation, distance vector calculation, dot product calculation, and then the improved interpolation method that introduces the texture adjustment factor is used to perform the final interpolation calculation noise, and then multiple layers of noise are superimposed. Here, the calculation steps are shown by taking the superposition of three layers of noise as an example:

[0072]

[0073] Among them, N i represents the Perlin noise of the i-th layer.

[0074] Finally, the final color set described in S102 is used for color mapping to obtain the final camouflage. The parameters involved in this process, including the texture adjustment factor, are left to be tuned.

[0075] S104. Reconstruct the underlying operator of the genetic algorithm according to the parameter tuning problem in the process of generating texture from noise, design a more comprehensive camouflage evaluation factor as the fitness of the genetic algorithm, and use it in the tuning process of key parameters of the process, wherein the optimization direction is specifically restricted for different types of camouflaged objects to adapt to the camouflage requirements of the camouflaged objects in the current target scene background, so that they can blend into this type of environmental background.

[0076] An improved genetic algorithm with redesigned selection operator, crossover operator and mutation operator is used, and new evaluation factors covering three dimensions of color matching, texture similarity and confusion degree are used as fitness to tune the parameters of the S103 process.

[0077] According to the tuning process, the camouflaged objects are classified, and the parameters that are more sensitive to each category are selected, that is, the key parameters when tuning the camouflaged objects of this category. In particular, range restrictions are imposed to prevent the camouflage texture from not matching the camouflage requirements of the camouflaged objects of this category, resulting in poor actual camouflage effect.

[0078] As an example, when the camouflage objects are personal equipment and uniforms, personal equipment needs to be effectively camouflaged at close range, so highly detailed and varied textures are required, so the following parameters are specifically limited:

[0079] Amplitude is between 0.5 and 2.0 because lower contrast is needed to avoid being too conspicuous at close range; Scale adjusts the level of detail and size of the pattern to suit the size of personal equipment and is between 10 and 100; α (the texture adjustment factor in the interpolation stage) is used to create delicate textures to effectively break up the human body outline at close range and is between 1 and 5.

[0080] For example, when the camouflaged object is medium- and long-range equipment such as armored vehicles, artillery systems, surface ships, and fixed-wing aircraft, the demand is more focused on improving the obfuscation effect, blending with the background environment, and reducing the possibility of being discovered by long-distance observers. Therefore, the following parameters must be specifically limited:

[0081] Scale needs to be medium-sized to accommodate observations at medium and long distances, between 50 and 500. Octaves needs to increase the number of noise layers to increase the complexity and details of the texture, between 4 and 8. Frequency is adjusted to match the environmental texture characteristics at medium and long distances, between 0.01 and 0.1.

[0082] S105, locally storing the tuned parameters together with the environmental image, evaluating and comparing with the locally stored image each time a new image is acquired in the target scene, determining whether to update the parameters or directly use the parameters to generate deformed camouflage, and then digitally generating a digital camouflage texture.

[0083] After each generation process is completed, the initial image and the tuned parameters will automatically determine whether to save them locally and overwrite and update the locally saved version to facilitate camouflage generation for the scene when no new image is provided.

[0084] S106, adjusting the layout of the digital camouflage texture module, the relationship between the main spots, the spot modification effect, etc. according to the segmentation attribute, the layer attribute and other camouflage related standard specifications.

[0085] The adjustment of segmentation requirements is mainly to prevent the overall direction, shape, size, and position distribution of the generated digital camouflage module from showing regularity, and to prevent the treatment of the interface of the camouflaged object surface from being inadequately standardized. The former is mainly achieved by adding small spots to the main spots of the digital camouflage module and adjusting the position and direction of the main spots of the digital camouflage module; the latter is mainly aimed at the main spots close to or touching the edge, extending them through the interface and crossing the boundary obliquely. Specifically, in implementation, when the distance between a main spot and the boundary is less than 10% of the horizontal or vertical length of the main spot, the main spot is extended to the adjacent surface, and the overall direction and the boundary angle are controlled to be as low as 30 degrees.

[0086] The adjustment of multi-layer requirements mainly lies in the addition of modified spots and the special treatment of large concave and convex parts of the camouflaged object. The former mainly adds inner spots inside the main spots with larger areas and peripheral spots around the outside. The inner spots use the brighter colors among the last three colors in the final color set and are different from the main spots, while the peripheral spots use the same color as the main spots; the latter mainly adds the main spots to the large concave and convex parts of the camouflaged object. The concave parts use the brighter colors among the last three colors in the final color set, and the convex parts use the darker colors among the last three colors in the final color set.

[0087] In an optional embodiment, if Figure 2 As shown, Figure 2 1 is a flow chart of color extraction provided by an embodiment of the present invention. This embodiment relates to how to use a multi-angle environment image to extract the main color set in the scene. Based on the above embodiment, the following steps are included:

[0088] S201, color clustering is performed on the five environment images respectively to obtain five color sets, and each color set retains four color elements according to their proportions in the image.

[0089] There are many pixel colors in each environment image. We first cluster them to obtain several colors that can represent the image. When each image color is clustered, the representative color mapped to it is consistent.

[0090] As an example, if the colors a, b, c clustered together in environment image 1 and the colors d, e, f, g clustered together in environment image 2 are relatively similar colors, then the representative colors mapped out after clustering of the two are the same.

[0091] Then, each color set is sorted from large to small according to the proportion of the color in the image before clustering, and the first 4 color members are retained.

[0092] S202, screening out colors that appear in all five color sets and putting them into a final color set.

[0093] Select the colors that are in all five color sets, that is, the colors that occupy the top four after the five environment image colors are clustered.

[0094] S203, using the high-frequency colors of the five color sets for neutralization, and calculating to obtain complementary colors, until the number of final color set members is 4.

[0095] If the five color sets are exactly the same, that is, a total of four colors are selected and put into the final color set, then there is no need to calculate the complementary colors. Otherwise, the complementary colors are obtained by neutralizing the high-frequency colors. For example, in S202, a total of two colors are selected and put into the final color set, that is, there are two colors that exist in the five color sets at the same time. In this case, two complementary colors need to be calculated. First, the first color in the five color sets is neutralized and averaged to obtain the first complementary color and put it into the final color set. Then, the second color in the five color sets is neutralized and averaged to obtain the second complementary color and put it into the final color set. At this point, there are four color members in the final color set.

[0096] S204, planning the specific use of the four color set members in camouflage according to the order of their entry.

[0097] As an example, assuming that the four color members are color x, y, z, and w in order, color x is the background color when generating camouflage, colors y and z are the colors of the main spots, and color w is a backup for adding additional modified spots later. In special cases such as inner spots and spots in concave and convex positions of camouflaged objects, colors y, z, and w may be set as the main spot color or the modified spot color depending on the relative light and dark conditions.

[0098] In an optional embodiment, if Figure 3 As shown, Figure 31 is a schematic diagram of a process of generating texture from noise and using an improved genetic algorithm to tune key parameters provided by an embodiment of the present invention. This embodiment relates to how to optimize the process of generating texture from noise using an improved genetic algorithm, and in particular, how to tune key process parameters. Based on the above embodiment, the following contents are included:

[0099] The entire texture generation process can be divided into Perlin noise calculation, noise superposition, and color mapping. There are many key parameters in the first two processes, which constitute the parameter set to be tuned. The process of improving the genetic algorithm for tuning is to regard each noise parameter set as an individual, and then let the individual be selected for crossover and mutation operations to continuously iterate. After a certain number of iterations, the excellent parameter individuals generated can be used to generate camouflage textures with better camouflage effects. The newly designed texture evaluation factor is the fitness used to evaluate the individual's excellence in this process. In summary, it allows the parameter set to iterate towards a higher score in the texture evaluation factor.

[0100] For the improved selection operator, the parameter individuals with texture evaluation factors greater than 0.8 for generating camouflage are considered as better individuals in advance, and a greater selection probability is given when calculating the selection probability, making it easier for good individuals to be selected. For the improved crossover operator and mutation operator, the new probability calculation algorithm allows individuals with fitness higher than the average level of the population to reduce the crossover and mutation probability as much as possible when the population is relatively dispersed, to ensure that individuals with good parameter sets can be retained and enter the next generation, so that the final result is better.

[0101] As an example, suppose there are many parameter set individuals in the current community, among which the parameters of parameter set individual 1 are Scale = 50.0, Amplitude = 3.5, Octaves = 3... The fitness is 0.81, the parameters of parameter set individual 2 are Scale = 100.0, Amplitude = 2.0, Octaves = 2... The fitness is 0.75, and the parameters of parameter set individual 3 are Scale = 40.0, Amplitude = 2.0, Octaves = 2... The fitness is 0.68, then parameter set individuals 1 and 2 are more likely to be selected than individual 3, and individual 1 is selected The probability is much greater than that of individual 2 and individual 3. At this time, if individual 1 and individual 2 are selected, when the crossover operation calculates the probability, on the one hand, the overall fitness of the two individuals is large, so there will be a positive bonus to the occurrence probability, but at the same time it will depend on the fitness of the current overall community. If the whole is dispersed, in order to retain the relatively excellent camouflage generation effect of the parameter set of individual 1, this crossover operation may still not occur; if the whole is concentrated, the two are more likely to have a crossover operation, which may produce, for example, a new parameter set individual: Scale = 50.0, Amplitude = 2.0, Octaves = 3... Its fitness is 0.84, that is, this process produces a better parameter set individual.

[0102] The present invention provides a method for generating digital camouflage based on noise. The above is only a preferred embodiment of the present invention. It should be noted that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.

Claims

1. A noise-based digital camouflage generation method, used to generate a camouflage digital camouflage suitable for a target scene according to a target scene, characterized in that: The following steps are involved: Step 1: In non-rainy or snowy weather with sufficient natural light, capture environmental images in the target scene from five directions: front, back, left, right, and bottom. Step 2: extracting the final main color set in the target scene from the five environment images for color distribution planning of the background, main spots, modified spots, etc. in the camouflage texture; Step 3, select multiple Perlin noises to generate camouflage textures and optimize the internal algorithm, further screen the parameters involved in the noise texture generation algorithm process, and obtain the key parameters of the process to be optimized; Step 4: Reconstruct the underlying operator of the genetic algorithm according to the parameter tuning problem in the process of generating texture from noise, design a more comprehensive camouflage evaluation factor as the fitness of the genetic algorithm, and use it in the tuning process of key parameters of the process, wherein the optimization direction is specifically restricted for different types of camouflaged objects to adapt to the camouflage requirements of the camouflaged objects in the current target scene background, so that they can be integrated into this type of environmental background; Step 5: Locally store the tuned parameters together with the environment image, evaluate and compare with the locally stored image each time a new image is acquired in the target scene, decide whether to update the parameters or directly use the parameters to generate deformed camouflage, and then digitally generate a digital camouflage texture module; Step 6: Adjust the layout of the digital camouflage texture module, the relationship between the main spots, the spot modification effect, etc. according to the national camouflage standards such as segmentation attributes and layer attributes.

2. A noise-based digital camouflage generation method according to claim 1, characterized in that: The specific process of extracting colors in step 2 to obtain the final main color set and plan the color distribution is: Using a clustering algorithm on the color data in the five environment images respectively, clustering pixels with similar colors into the same category, and outputting the representative color of the category, which is recorded as one color; Similarly, the first 4 colors of each environment image are obtained and stored in color set C. Five color sets C are obtained from five images. i ,i∈{1,2,3,4,5},c mn Represents the color set C obtained by the mth image m The color with the nth largest proportion; If for c mn ∈C i , then c mn Put it into the final color set G, and finally get the final color set G with x elements. If x<4, take The color g n The final color set G is also stored, so that the final color set G contains 4 color elements as the final color set of the target scene; The first color in the final color set G is used as the main background color of the camouflage texture, the second and third colors are used as optional colors for generating the main spots, and the fourth color is reserved as a color for adding additional modification spots.

3. The noise-based digital camouflage generation method according to claim 1, characterized in that: The specific steps of optimizing and adjusting the internal algorithm in step 3 to generate camouflage texture by superimposing multiple Perlin noises and screening key parameters in the process are as follows: After the early steps of noise calculation such as grid network definition, gradient vector allocation, distance vector calculation, and dot product calculation, when the dot product values ​​of adjacent grid points are combined through an interpolation function, the present invention introduces a texture adjustment factor α in conventional bilinear interpolation. The noise calculation of the conventional bilinear interpolation is: N(x,y)=lerp(u,lerp(v,dot i,j ,dot i+1,j ),lerp(v,dot i,j+1 ,dot i+1,j+1 )) The improved interpolation process after introducing the texture adjustment factor is: Where lerp(t,a,b) represents the linear interpolation function, dot i,j 、dot i+1,j 、dot i,j+1 、dot i+1,j+1 Respectively represent the dot product between the gradient vector of the four vertices, namely the bottom left vertex, the bottom right vertex, the top left vertex, and the top right vertex, and the distance vector from the vertex to the interpolation point (x, y). Respectively represent the noise values ​​obtained after adjusting the dot product values ​​of each vertex, u and v are the smoothing function values ​​in the x and y directions based on the input point coordinates, and the smoothing function uses the improved 5th-order Hermite smoothing function to ensure better interpolation smoothness; By adjusting the value of α, you can create richer and more delicate effects in different texture areas, thereby better simulating the variability and complexity of camouflage textures in nature. α will also be an important parameter to be tuned; Then, by superimposing multiple layers of Perlin noise of different frequencies and amplitudes, complex and diverse textures are generated, and after the noise superposition is completed, color mapping is performed according to the final color distribution plan described in claim 2 to generate a final camouflage texture; The final parameters to be tuned are: the amplitude attenuation rate between different frequency layers in the noise texture, i.e., persistence, the texture adjustment factor α in the improved interpolation algorithm, the size of the features in the noise texture, i.e., the noise texture detail ratio Scale, frequency Frequency, amplitude Amplitude, noise layer coefficient Octaves, lacunarity, offset offset, noise gain gain, threshold Threshold for binarization of noise values, Maximum value for limiting the maximum range of noise values, and Minimum value for limiting the minimum range of noise values.

4. The noise-based digital camouflage generation method according to claim 1, characterized in that: The improved genetic algorithm applied to this problem in step 4 is specifically as follows: The fitness of the improved genetic algorithm is a new camouflage evaluation factor, and the calculation method is: Calculating the histogram of the final camouflage texture image and the environment image, and comparing the similarity between them using the Bhattacharyya coefficient, wherein the larger the coefficient is, the higher the similarity between the two images in color is; Using GLCM to evaluate the texture similarity between the final camouflage texture image and the environment image. GLCM is a statistical method for analyzing image texture features. It describes the texture features of an image by calculating the grayscale difference of pixels in different directions. Using the GLCM comparison function to calculate the similarity between two groups of GLCMs, the larger the value, the higher the texture similarity between the two images. Use Canny edge detection to evaluate the confusion effect of the final camouflage texture image and the environment image. Canny edge detection is a commonly used edge detection algorithm that can detect obvious edges in an image. The two images are subjected to Canny edge detection respectively, and then their edge information is compared. The smaller this value is, the better the confusion effect of the camouflage texture image and the environment background image on the edge is; The scores of the three dimensions are averaged to obtain a comprehensive score, and the range is controlled to be between 0 and 1, which is used as a camouflage evaluation factor, that is, the fitness of the improved genetic algorithm; The selection operator of the improved genetic algorithm is improved based on the ordinary roulette selection algorithm, and the new evaluation factor is used as the fitness. Through a large number of experiments, it is found that the texture evaluation factor with excellent effect is usually above 0.8, so the individuals with fitness higher than 0.8 are marked as excellent individuals, and the probability of the excellent individuals is increased to two times, making it easier to be selected and pass on the excellent features, that is, the probability P of each chromosome being selected is calculated each time i , there are M chromosomes in total, f i is the fitness of the chromosome, P i It can be expressed as The crossover method selected by the crossover operator of the improved genetic algorithm is single-point crossover, and the mutation method selected by the mutation operator is single-point mutation. Both the crossover probability and the mutation probability are dynamically adjusted according to the population state: Crossover probability: Where P c is the crossover probability, P c_max is the maximum crossover probability in the current population, f′ is the larger fitness value of the two parent individuals in the crossover operation, and f max is the maximum fitness value in the entire population, f min is the minimum fitness value. In this application, the fitness function is accuracy, so f∈[0,1],f max -f min ∈[0,1]; Probability of mutation: Where P m is the mutation probability, P m_max is the current maximum mutation probability, f is the fitness value of the individual, and the rest is the same as the crossover probability part above.

5. The noise-based digital camouflage generation method according to claim 1, characterized in that: The specific content of the tuning range limit for different types of camouflage objects in step 4 is as follows: For the uniforms, helmets, carrying gear, weapons and other personal equipment and uniform camouflage of relevant personnel, the range restrictions of the parameters such as Amplitude, Scale, α, etc. in claim 3 are added, specifically: Amplitude is between 0.5 and 2.0, Scale is between 10 and 100, and α is between 1 and 5; For the camouflage of medium and long-range equipment such as armored vehicles, non-armored support vehicles, artillery systems, surface ships, fixed-wing aircraft, etc., add the range limits of the parameters such as Scale, Octaves, and Frequency as stated in claim 3, specifically: Scale is between 50 and 500, Octaves is between 4 and 8, and Frequency is between 0.01 and 0.1; For the camouflage of fixed military facilities such as command posts, fortifications, logistics facilities, bridges and important transportation nodes, the range limits of the parameters such as Scale, Octaves, α, etc. in claim 3 are added, specifically: Scale is between 100 and 1000, Octaves is between 6 and 10, and α is between 2 and 6; For submarine and other underwater equipment camouflage, add the range limits of parameters such as Scale, Octaves, and Persistence in claim 3, specifically: Scale is between 20 and 200, Octaves is between 5 and 9, and Persistence is between 0.4 and 0.7; For tactical deception camouflage such as decoys and false targets, the range limits of parameters such as Octaves and α mentioned in claim 3 are added, specifically: Octaves is between 3 and 7, and α is between 3 and 8; The above ranges are all closed ranges.

6. The noise-based digital camouflage generation method according to claim 1, characterized in that: The specific steps of parameter updating and specific application in the actual scenario in step 5 are as follows: Input a new environment image collected from the target scene, use the first two evaluation dimensions of the evaluation factor described in claim 4, and perform color similarity evaluation and texture matching evaluation with the stored image of the scene, take the average of the two, and the final result ranges from 0 to 1; if there is no stored image, that is, it is the first time to run under the target scene, then the evaluation result returned is 0; When the evaluation result is less than 0.5, re-tuning is performed using the improved genetic algorithm of claim 4 to obtain a new parameter set, the new parameter set is used to overwrite and update the parameter set of the target scene, and the newly provided image of the current environment is overwritten to update the locally stored image of the target scene; After the evaluation is completed, the parameter set is used to generate noise texture according to the specific steps in claim 3 to produce a final camouflage texture, that is, a deformed camouflage texture, and then a grid of appropriate size is constructed based on the deformed camouflage. When the pixels of the main pattern color in a certain grid are more than 50%, the grid is set to the main pattern color, and finally a digital camouflage texture is obtained.

7. The noise-based digital camouflage generation method according to claim 1, characterized in that: In step 6, the digital camouflage module is adjusted according to the relevant camouflage standards, and the specific contents are as follows: According to the segmentation requirements, if the distance between a main spot on a certain face of the camouflaged object and the edge is less than 10% of the width of the face in the direction perpendicular to the edge, the whole pattern is extended in the direction perpendicular to the edge and rotated clockwise until the overall direction of the spot pattern forms an angle of 30 degrees with the edge; According to the multi-level requirements, for the main spot on a certain surface of the camouflaged object, calculate its maximum circumscribed rectangular area. If it is larger than 15% of the area of ​​the surface where the main spot is located, add peripheral modification spots of the same color to the outside and add inner modification spots of different colors to the inside; for the concave part of the camouflaged object, if the area exceeds 20% of the area of ​​the surface where the concave part is located, add the main spot to the concave part and use the brighter color member of the final color set described in claim 2; for the convex part of the camouflaged object, if the area exceeds 20% of the area of ​​the surface where the concave part is located, add the main spot to the convex part and use the darker color member of the final color set described in claim 2; For an independent main spot on the camouflaged object, if there is a continuous rectangular area in the horizontal, vertical and oblique directions of the maximum circumscribed rectangle of the main spot without any main spots and modified spots, and the area of ​​the blank rectangular area exceeds 80% of the area of ​​the maximum circumscribed rectangle of the main spot, then other main spots and corresponding modified spots in a staggered, juxtaposed or other positional relationship are added to the blank rectangular area, that is, next to the original independent main spot.

Images