Camouflage generation method and device

Abstract

The invention discloses a camouflage generation method and equipment, and belongs to the field of camouflage. According to the invention, an embedded computing platform is utilized to drive a multi-channel camera to collect a background environment, and image splicing processing is carried out to obtain a data source of a camouflage image. By adopting data filtering operation based on image feature clustering analysis, the number of redundant alternative data blocks is reduced, and computing resources are saved. By adopting the comprehensive evaluation indexes of color similarity and SSIM structure similarity in the HSV space, the filling blocks conforming to the optimal matching characteristics at the color and texture levels are screened out, and the quality of the generated image is improved. And the filling sequence is further optimized by adopting a priority function for increasing color consistency factors, so that the quality of the generated image is improved. By adopting the edge fuzzy processing of the to-be-filled region, the boundary splitting problem of the generated region and the known background region is improved. The display device is utilized to present and display the generated camouflage image, so that stealth camouflage in the environment is realized.

Description

Technical Field

[0001] The present invention relates to a camouflage pattern generation method and device, and in particular to a camouflage pattern image generation method based on an image restoration method and a special generation device, belonging to the field of camouflage patterns. Background Art

[0002] Before executing a mission, drones must accurately detect and identify ground targets. Effective counter-reconnaissance technology is crucial for ground targets, making camouflage technology a crucial component of visible light camouflage. This technology allows ground-based equipment to better conceal itself on the battlefield, giving them an advantage in intense combat situations. To address diverse and complex operational scenarios, the design and development of equipment capable of adaptively changing camouflage patterns has significant practical application value.

[0003] Existing camouflage pattern generation methods and equipment often rely on independent, step-by-step design procedures for color, texture, and patch. These methods are based on the characteristics of human vision, and their design metrics are subjective and understandable. However, they perform poorly against algorithmic reconnaissance by unmanned equipment, and the resulting camouflage patterns often lack texture and color integration with the surrounding background. To design camouflage images with improved camouflage performance, we need to consider design metrics more comprehensively and employ other innovative methods to achieve better camouflage performance. Summary of the Invention

[0004] In order to solve the problems of poor generation effect of existing camouflage pattern generation methods, lack of global optimization of generated camouflage, and insufficient matching between color distribution and texture structure, the purpose of the present invention is to provide a camouflage pattern generation method and device, which realizes camouflage pattern image generation based on the Criminisi image restoration method, solves the problem of poor camouflage pattern generation effect by increasing data filtering, optimizing the best filling block evaluation index, improving the priority function and boundary blurring improvement measures, and achieves the global optimal solution of color and texture features of generated camouflage. The output camouflage image has a higher similarity with the environment, thereby improving the camouflage pattern effect.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] The present invention discloses a camouflage pattern generation method, which uses an embedded computing platform to drive a multi-channel camera to capture the background environment and perform image splicing processing to obtain the data source of the camouflage image, thereby ensuring the data authenticity of the camouflage image. By adopting a data filtering operation based on image feature clustering analysis, the number of redundant alternative data blocks is reduced, computing resources are saved, and the iterative operation efficiency of the generation method is improved. By adopting the comprehensive evaluation index of color similarity and SSIM structural similarity in HSV space, the filling blocks that meet the best matching characteristics at the color and texture levels are screened out to improve the quality of the generated image. By adopting a priority function that increases the color consistency factor, the filling order is further optimized, thereby improving the quality of the generated image. By adopting the edge fuzzy processing of the area to be filled, the boundary separation problem between the generated area and the known background area is improved. By utilizing a display device, the generated camouflage image is presented and displayed, thereby achieving stealth camouflage in the environment.

[0007] The present invention discloses a camouflage pattern generation method, comprising the following steps:

[0008] S1. Use the embedded computing platform to drive the multi-channel camera module to collect multiple background images of the actual working scene, stitch the images according to the relative positions of the cameras in the device, and plan the area to be repaired.

[0009] S2. Using the embedded computing platform to perform data preprocessing on the background image after the splicing process, performing data filtering operations based on image feature clustering analysis, and calculating and obtaining a spare data filling block.

[0010] S3. Obtain edge pixels of the area to be repaired based on the embedded platform, calculate the repair priority of the edge pixels based on the priority function of the fused color consistency factor, and confirm the pixel area to be repaired first according to the priority.

[0011] S4. Based on the embedded computing platform, from all the candidate data filling blocks generated in step S2, the comprehensive evaluation index integrating HSV space color similarity and SSIM structural similarity is used to screen out the filling blocks obtained in step S3 that best match the priority repair area, and place the filling blocks in the priority area to be repaired.

[0012] S5. Based on the embedded computing platform, use the edge extraction algorithm to detect whether the image restoration of the area to be restored is completed. If the image restoration is completed, execute step S6. If not, execute steps S3, S4 and S5 in a loop.

[0013] S6. Based on the embedded computing platform, the generated camouflage image is transmitted to the display device for display, thereby realizing stealth camouflage in the working background environment.

[0014] The specific implementation method of S2 is:

[0015] S20. Based on the embedded computing platform, read the height and width of the known region of the image, set the basic filling block size parameter PatchSize, set the sliding step parameter Step, and calculate the number of preliminary candidate blocks. Through a double loop, traverse the known region of the image, extract candidate blocks of PatchSize × PatchSize at intervals of Step, and temporarily store all candidate blocks.

[0016] S21. Load the pre-trained ResNet-50 model, scale all candidate blocks to a certain size, input them into the ResNet-50 model, extract the multi-dimensional feature vector, and perform normalization.

[0017] S22. Execute the K-means clustering algorithm on the standardized feature vector obtained in step S21, specify the number of output clusters num_clusters, and obtain cluster labels for all candidate blocks. Select a specified number of candidate blocks from each cluster based on the Euclidean distance between each candidate block and its cluster center.

[0018] S23: All the candidate blocks after screening are stored in the memory for use in the subsequent step S3, so as to effectively improve the operation efficiency of the algorithm.

[0019] The specific implementation method of S3 is:

[0020] S30 , calculating the intersection of the known area image and the position area image of the background image to obtain edge pixels.

[0021] S31. Calculate the confidence of edge pixels. The confidence term is the ratio of known pixels in the current priority repair area to all pixels. The calculation method is shown in formula (1).

[0022]

[0023] In the formula, C(q) is the confidence term, PatchSize 2 Indicates the total number of pixels in the current repair block, including the known part and the part to be repaired. q represents the set of currently inpainted regions, and A represents the known regions in the image. The confidence items for all pixels in the known regions are initially set to 1, while the initial value for the pixels to be inpainted is 0. C(k) represents the known regions in the region, so the confidence item determines the reliability based on the number of known pixels in the region.

[0024] S32. Calculate the data item of edge pixels. The data item is based on the grayscale gradient amplitude of the image. The larger the gradient value, the more useful information it represents. The gradient calculation is divided into x and y directions. Convolution is used for calculation. The horizontal gradient G x , vertical gradient Gy The calculation methods of the total gradient D(q) are shown in equations (2), (3), and (4) respectively.

[0025]

[0026] Where, ψ q Indicates the current repair area. The data item reflects the edge structure information. The larger the value, the greater the impact of the area on visual continuity, which means that the area should be repaired first.

[0027] S33. Calculate the color consistency factor. The color consistency factor refers to the difference between the average color of the current repair block and the global average color. The smaller the color difference, the smaller the error caused by the repair of the current repair block. The calculation method is shown in formula (5).

[0028]

[0029] Where S(q) represents the color consistency factor, is the average color of the current repair block, represents the global average color of the image, σ 2 represents the Gaussian smoothing parameter.

[0030] S34. Calculate the priority function value. The calculation factors are derived from steps S31, S32, and S33. The calculation method is shown in formula (6).

[0031] P(q)=C(q)×D(q)×S(q) (6)

[0032] S35. Filter out the maximum value and its corresponding edge pixel from all calculated priority function values.

[0033] The specific implementation method of S4 is:

[0034] S40 , obtaining a known portion of the area to be repaired by calculating the intersection of the known pixel area and the current area to be repaired.

[0035] S41. Calculate the SSIM values and HSV space color difference values of the known pixel area and the alternative filling block in the area to be repaired, and calculate the joint index of color and structure based on the above two parameters. The calculation method is shown in formula (7).

[0036]

[0037] Where S C and S S Represents color similarity and structural similarity respectively. Calculate the evaluation results of the current block to be repaired and all candidate filling blocks, and select the most similar area as the filling block.

[0038] S42: Overwrite the area to be repaired with the value of the filling block to achieve one-time filling of the area to be repaired.

[0039] The present invention discloses a camouflage pattern generating device, which is used to implement the camouflage pattern generating method. The camouflage pattern generating device includes an embedded computing platform, a multi-channel camera module, a power supply module and a display device.

[0040] The embedded computing platform is a SoC platform, which is used to drive a multi-channel camera module, to process collected images, to run the algorithm program of the camouflage pattern generation method, and to drive a display device.

[0041] The multi-channel camera is used to capture background environment images from multiple perspectives.

[0042] The power supply module is used to provide power to the embedded platform, multi-channel camera and display device.

[0043] The display device is used to receive image data from the embedded platform and emit electromagnetic spectrum to the outside to realize the presentation of camouflage images.

[0044] Beneficial effects:

[0045] 1. The present invention discloses a camouflage pattern generation method and device. By utilizing an embedded computing platform to drive a multi-channel camera to capture the background environment and perform image stitching processing, the data source of the camouflage image is obtained, thereby ensuring the data authenticity of the camouflage image. Furthermore, data filtering operations are performed on candidate data blocks. By adopting data filtering operations based on image feature clustering analysis, the number of redundant candidate blocks is reduced, storage resources are saved, and the iterative running time of a camouflage pattern generation method is shortened, thereby improving the efficiency of camouflage image generation.

[0046] 2. The present invention discloses a camouflage pattern generation method and device, which adds a color consistency factor to the priority function used to plan the filling order, improves the filling order of the boundaries of the area to be repaired, and thus improves the quality of the generated image.

[0047] 3. The present invention discloses a camouflage pattern generation method and device, which uses a comprehensive evaluation index of color similarity and SSIM structural similarity in HSV space to screen out filling blocks that meet the best matching characteristics at the color and texture levels. The index used to evaluate the best filling blocks is replaced with a comprehensive evaluation index based on HSV color similarity and SSIM structural similarity, thereby improving the quality of the generated image and enhancing the camouflage effect of the generated area.

[0048] 4. The present invention discloses a camouflage pattern generation method and device, which adds a fuzzy processing mechanism to the boundary of the repair area, effectively improving the boundary separation problem between the generated area and the known background area, and improving the camouflage effect of the generated image.

[0049] 5. This invention discloses a camouflage pattern generation method and device that utilizes embedded computing devices for algorithm deployment and a portable power module for power supply, effectively expanding operational scenarios. The generated camouflage pattern is displayed on a display device, achieving stealth and camouflage in the environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a camouflage pattern generation method;

[0051] Figure 2 is the image to be repaired in Experiment A after splicing;

[0052] Figure 3 is the image to be repaired in experiment B after splicing;

[0053] Figure 4 This is the camouflaged image generated by experiment A based on image restoration;

[0054] Figure 5 This is the camouflaged image generated by experiment B based on image restoration;

[0055] Figure 6 is the original background image of Experiment A;

[0056] Figure 7 is the original background image of Experiment B.

[0057] Figure 8 The relationship between the working platform and the camouflage image generation equipment;

[0058] Figure 9 The structural composition of the camouflage image generation device. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments and the accompanying drawings.

[0060] like Figure 1 As shown, the camouflage pattern generation method disclosed in this embodiment has the following specific implementation steps:

[0061] Step 1: Based on NVIDIA's Jetson Xavier NX embedded computing platform, under the Linux system, drive a multi-channel USB driver-free camera to shoot and record multiple background images with a size of 640×480, read the background images, and stitch them according to the relative spatial position of the cameras to obtain images containing the area to be filled and the known area. This includes two experimental verifications, such as Figure 2 and Figure 3 As shown in the figure, both experiments set PatchSize to 81, Step to 20, and the number of clusters num_clusters to 100. The known regions of the background image were segmented using a sliding step size to create candidate blocks. The clustering algorithm was then applied to all candidate blocks, classifying them into 100 categories. The 100 image blocks at the cluster centers were selected as the final candidate blocks. That is, the candidate blocks consisted of 100 image blocks with typical background features, all with a size of 81×81.

[0062] Step 2: Calculate the priority value of the edge of the area to be repaired according to the calculation method of the priority function including the confidence item, data item and color consistency factor, and filter out the edge pixel with the highest priority value. The area with the edge pixel as the center and a side length of 81 is the current area to be repaired.

[0063] Step 3: Calculate the comprehensive evaluation index of HSV space color similarity and SSIM structure similarity for the current area to be repaired and the 100 candidate blocks calculated in step 1 above, and select the best matching candidate block and fill it into the current area to be repaired.

[0064] Step 4: Update the edge of the area to be repaired and check whether there is an unrepaired area. If there is, repeat steps 2 and 3. If not, it means that the area to be repaired has been repaired and the generated camouflage image is obtained. The generated images of the area to be repaired in the two experiments are as follows: Figure 4 and Figure 5 As shown in the figure, the original images of the two experiments are as follows Figure 6 and Figure 7 shown.

[0065] Step 5: Display the camouflage image generated in step 4 on a display device, achieving camouflage and stealth protection for the platform in an actual operational scenario. The camouflage pattern design method and apparatus described in this embodiment can effectively improve the camouflage effect of camouflage patterns. In the analysis of this embodiment, the camouflage image generated through image restoration has a high degree of color, texture, and structural similarity to the original background.

[0066] The present invention discloses a camouflage pattern generating device, which is used to implement the camouflage pattern generating method. The camouflage pattern generating device includes an embedded computing platform, a multi-channel camera module, a power supply module and a display device. The camouflage pattern generating device disclosed in this embodiment is installed in a working platform, such as Figure 8 As shown, the embodiment disclosed in this invention is a camouflage pattern generating device, as shown in FIG. Figure 9 shown.

[0067] The embedded computing platform is a SoC platform with a certain computing power, which is used to drive a multi-channel camera module, to process the collected images, to implement the computational implementation of a camouflage pattern generation method, and to drive a display device.

[0068] The multi-channel camera is used to capture background environment images from multiple perspectives.

[0069] The power supply module is used to provide power to the embedded platform, multi-channel camera and display device.

[0070] The display device is used to receive image data from the embedded platform and emit electromagnetic spectrum to the outside to realize the presentation of camouflage images.

[0071] The above specific description further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A camouflage pattern generation method, characterized by: The following steps are included: S1. Use the embedded computing platform to drive the multi-channel camera module to collect multiple background images of the actual working scene, stitch the images according to the relative positions of the cameras in the device, and plan the area to be repaired; S2. Using an embedded computing platform to pre-process the background image after the stitching process, perform data filtering operations based on image feature clustering analysis, and calculate and obtain a spare data filling block; S3. Obtain edge pixels of the area to be repaired based on the embedded platform, calculate the repair priority of the edge pixels based on the priority function of the fused color consistency factor, and determine the pixel area to be repaired first according to the priority; S4. Based on the embedded computing platform, from all candidate data filling blocks generated in step S2, a comprehensive evaluation index combining HSV spatial color similarity and SSIM structural similarity is used to screen out the filling blocks obtained in step S3 that best match the priority repair area, and place the filling blocks in the priority repair area. S5. Based on the embedded computing platform, use the edge extraction algorithm to detect whether the image restoration of the area to be restored is completed. If the image restoration is completed, execute step S6. If not, execute steps S3, S4 and S5 in a loop. S6. Based on the embedded computing platform, the generated camouflage image is transmitted to the display device for display, thereby realizing stealth camouflage in the working background environment.

2. The camouflage pattern generation method according to claim 1, wherein: The specific implementation method of S2 is: S20, based on the embedded computing platform, read the height and width of the known area of the image, set the basic filling block size parameter PatchSize, set the sliding step parameter Step, and calculate the number of preliminary candidate blocks; traverse the known area image through a double loop, extract candidate blocks of PatchSize × PatchSize at intervals of Step, and temporarily store all candidate blocks; S21. Load the pre-trained ResNet-50 model, scale all candidate blocks to a certain size, input them into the ResNet-50 model, extract the multi-dimensional feature vector, and perform normalization. S22. Execute the K-means clustering algorithm on the standardized feature vector obtained in step S21, specify the number of output clusters num_clusters, and obtain cluster labels for all candidate blocks; select a specified number of candidate blocks from each cluster based on the Euclidean distance between each candidate block and its cluster center; S23. Store all the selected candidate blocks into the memory.

3. The camouflage pattern generation method according to claim 2, wherein: The specific implementation method of S3 is: S30, calculating the intersection of the known area image and the position area image of the background image to obtain edge pixels; S31, calculating the confidence of edge pixels, where the confidence term is the ratio of known pixels in the current priority repair area to all pixels, and the calculation method is shown in formula (1); In the formula, C(q) is the confidence term, PatchSize 2 Indicates the total number of pixels in the current repair block, including the known part and the part to be repaired; ψ q Represents the current repair area set, A represents the known area in the image; the initial value of the pixel confidence item of all known areas is 1, and the initial value of the item to be repaired is 0; C(k) represents the known part of the area, so the confidence item determines the credibility according to the number of known pixels in the area; S32. Calculate the data item of edge pixels. The data item is based on the grayscale gradient amplitude of the image. The larger the gradient value, the more useful information it represents. The gradient calculation is divided into x and y directions. Convolution is used for calculation. The horizontal gradient G x , vertical gradient G y The calculation methods of the total gradient D(q) are shown in formulas (2)(3)(4) respectively; Where, ψ q Indicates the current repair area. The data item reflects the edge structure information. The larger the value, the greater the impact of the area on visual continuity, which means that the area should be repaired first. S33. Calculate the color consistency factor. The color consistency factor refers to the difference between the average color of the current repair block and the global average color. The smaller the color difference, the smaller the error caused by the repair of the current repair block. The calculation method is shown in formula (5); Where S(q) represents the color consistency factor, is the average color of the current repair block, represents the global average color of the image, σ 2 represents the Gaussian smoothing parameter; S34, calculate the priority function value, the calculation factors are derived from steps S31, S32 and S33, and the calculation method is shown in formula (6); P(q)=C(q)×D(q)×S(q) (6) S35. Filter out the maximum value and its corresponding edge pixel from all calculated priority function values.

4. The camouflage pattern generation method according to claim 3, wherein: The specific implementation method of S4 is: S40, obtaining a known portion of the area to be repaired by calculating the intersection of the known pixel area and the current area to be repaired; S41, calculating the SSIM value and HSV space color difference value of the known pixel area and the candidate filling block in the repair area, and calculating the joint index of color and structure based on the above two parameters, the calculation method is shown in formula (7); Where S C and S S Represents color similarity and structural similarity respectively; calculates the evaluation results of the current block to be repaired and all candidate filling blocks, and selects the most similar area as the filling block; S42: Overwrite the area to be repaired with the value of the filling block to achieve one-time filling of the area to be repaired.

5. The camouflage pattern generation method according to claim 1, wherein: By utilizing an embedded computing platform to drive a multi-channel camera to capture the background environment and perform image stitching processing, the data source of the camouflaged image is obtained to ensure the data authenticity of the camouflaged image; data filtering operations are performed on the alternative data blocks, and by adopting data filtering operations based on image feature clustering analysis, the number of redundant alternative blocks is reduced and storage resources are saved.

6. The camouflage pattern generation method according to claim 1, wherein: By adopting the comprehensive evaluation index of color similarity and SSIM structural similarity in HSV space, we screen out the filling blocks that meet the best matching characteristics at the color and texture levels. The index used to evaluate the best filling blocks is replaced with a comprehensive evaluation index based on HSV color similarity and SSIM structural similarity to improve the quality of the generated image.

7. The camouflage pattern generation method according to claim 1, characterized in that: Use embedded computing devices to deploy camouflage pattern generation algorithms and power them with portable power modules to expand working scenarios.

8. A camouflage pattern generating device, used to implement a camouflage pattern generating method according to claim 1, 2, 3 or 4, characterized in that: Including embedded computing platform, multi-channel camera module, power supply module and display device; The embedded computing platform is a SoC platform, which is used to drive a multi-channel camera module, process collected images, run the algorithm program of the camouflage pattern generation method, and drive a display device; The multi-channel camera is used to capture background environment images from multiple perspectives; The power supply module is used to power the embedded platform, multi-channel camera and display device; The display device is used to receive image data from the embedded platform and emit electromagnetic spectrum to the outside to realize the presentation of camouflage images.

Images