A digital camouflage effect evaluation method

Abstract

The application discloses a digital camouflage effect evaluation method, and belongs to the field of digital camouflage. The method comprises the following steps: collecting a background image of a working scene by using a flight shooting device, and obtaining a digital camouflage image matched with the working background; processing and calculating the background image and the digital camouflage image by using a computing platform; selecting multiple parameter indexes including a gray scale similarity, a color similarity, a texture similarity, a patch size similarity, an edge similarity and an edge fusion degree; fusing the indexes into a comprehensive evaluation index by assigning weights to the indexes; evaluating the digital camouflage effect according to the comprehensive evaluation index, realizing quantitative analysis and efficient evaluation; and guiding the design and optimization of the digital camouflage according to the quantitative analysis results of the sub-parameters.

Description

Technical Field

[0001] This invention relates to a method for evaluating the effect of digital camouflage, and more particularly to a method for evaluating the effect of digital camouflage that integrates the overall and local features of digital camouflage images and can be used to guide the generation of camouflage images, belonging to the field of digital camouflage. Background Technology

[0002] For ground targets, effective counter-reconnaissance technology is particularly crucial, making digital camouflage technology a vital component in the field of visible light camouflage. Through digital camouflage, ground equipment can better conceal itself on the battlefield, thus gaining an advantage in intense combat. Accurately evaluating the camouflage effectiveness of digital camouflage is of significant practical value in designing and developing digital camouflage that can flexibly adapt to various operational scenarios.

[0003] In existing methods for evaluating the camouflage effect of digital camouflage images, the comprehensive evaluation index mainly relies on global feature parameters such as brightness similarity, color similarity, and texture similarity between the camouflage and background images. These methods regard image similarity as a standard for good camouflage effect. However, they are insufficient in considering local features such as the interface edge information between the camouflage and background images, and the independent contrast characteristics of existing methods are prone to causing an imbalance in feature weight allocation, further affecting the accuracy of digital camouflage evaluation. To optimize these evaluation methods, we need to consider the relationship between image similarity and camouflage effect more comprehensively and incorporate local feature parameters into the evaluation process to achieve a more accurate evaluation of camouflage effect. Summary of the Invention

[0004] Existing methods for evaluating the camouflage effect of digital camouflage suffer from insufficient evaluation indicators and low accuracy. This invention aims to provide a new method for evaluating the camouflage effect of digital camouflage. Based on the needs of actual application scenarios, it flexibly selects multiple parameter indicators, including grayscale similarity, color similarity, texture similarity, target patch size similarity, edge similarity, and edge blending degree. By assigning weights to these indicators, they are integrated into a comprehensive evaluation index, improving the accuracy of digital camouflage evaluation and contributing to the optimized design and engineering application of digital camouflage. This invention relies on a computing platform and can be deployed on edge computing devices.

[0005] The specific technical solution adopted in this invention is as follows:

[0006] This invention discloses a method for evaluating the camouflage effect of digital camouflage, comprising the following steps:

[0007] S1. Use aerial photography equipment to collect background images of the actual work scene, and at the same time obtain digital camouflage images that match the background; the aerial photography equipment includes drones, satellites, unmanned airships, helicopters, airplanes and parachutes.

[0008] S2. Calculate the length and width of the digital camouflage image using a computing platform; determine the target camouflage position coordinates of the digital camouflage image in the original background image; use these position coordinates as the camouflage center to divide the original background image into a nine-square grid region, with the length and width of each square grid consistent with the size of the camouflage image; this nine-square grid region is the background image that actually participates in the feature extraction calculation; the computing platform includes x86 CPU platform, ARM SOC platform, NVIDIA GPU and its Jetson edge computing processor, etc.

[0009] S3. The processed background image and digital camouflage image are stored in a computing platform. The computing platform contains multiple evaluation parameter models for evaluating the digital camouflage effect as described in this invention, specifically including the grayscale similarity S between the background image and the digital camouflage image. G Color similarity S C Texture similarity S T Patch size similarity S P Edge similarity S E Edge blending degree S H The above indicators are combined according to certain weighting parameters:

[0010] S = a × S G +b×S C +c×S T +d×S P +e×S E +f×S H (1)

[0011] In the formula, a, b, c, d, e, and f represent the weight parameters of grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge fusion degree, respectively.

[0012] The digital camouflage effect is evaluated based on the comprehensive evaluation index, and the evaluation result of digital camouflage is obtained. The output range of the comprehensive evaluation index is (0,1). The larger the value, the better the camouflage performance.

[0013] It also includes step S4: Based on the digital camouflage evaluation results obtained in step S3, it is possible to extract various feature information, including grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge blending degree, from the background image and the digital camouflage image; it is possible to evaluate the camouflage effect of the digital camouflage image in the background image using a comprehensive evaluation index of various feature information; and it is possible to provide effective and objective guidance for the design of digital camouflage based on the evaluation data of various sub-parameters such as grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge blending degree, and solve related engineering problems.

[0014] The specific implementation method of S3 is as follows:

[0015] S30. Select evaluation parameters for assessing the digital camouflage effect. The evaluation parameters include the grayscale similarity S between the background image and the digital camouflage image. G Color similarity S C Texture similarity S T Patch size similarity S P Edge similarity S E Edge blending degree S H .

[0016] S31, the grayscale similarity S G We obtain it from the following formula:

[0017] The background image and the digital camouflage image are scaled to the same size and converted to grayscale images. Then, the two images are scaled to the same size, and their respective grayscale histograms are constructed. The intersection data of the two grayscale histograms is calculated. Then, the intersection data of all grayscale levels is proportionally calculated with the sum of the pixels of the image to obtain the grayscale similarity as shown in Equation (2).

[0018]

[0019] In the formula, N G-t (i) represents the number of pixels with the i-th gray level value in the gray-level histogram of the digital camouflage image, N G-b (i) represents the number of pixels with the i-th gray level in the gray-level histogram of the background image. i represents the gray level, with a maximum output of 256 levels.

[0020] Gray-scale similarity S G The output range is (0,1). The larger the value, the more similar the grayscale features of the digital camouflage image and the background image are.

[0021] S32, the color similarity S C It is obtained through the following method:

[0022] The background image and the digital camouflage image are scaled to the same size and converted to HSV space. The colors are further quantized according to actual needs, and the quantization formulas are shown in equations (3), (4), and (5).

[0023]

[0024] In equation (3), h represents the hue value before quantization, H represents the hue value after quantization, and N... H -1 is the maximum output value of the quantized hue, C p This represents the hue interval bias value before quantization. In equation (4), v represents the saturation value before quantization, V represents the saturation value after quantization, and N represents the saturation value after quantization. V -1 represents the maximum output value of saturation after quantization, k1, k2, ... This is the interval boundary value of saturation before quantization. In equation (5), s represents the brightness value before quantization, S represents the brightness value after quantization, and N... S -1 represents the maximum output value of the quantized brightness, k1, k2, ... These are the boundary values ​​of each interval. The quantization interval boundary values ​​of V and S are consistent. The color components are synthesized into a one-dimensional vector according to equation (6).

[0025] L=9H+3S+V (6)

[0026] The color range L becomes 9N. H +3N V +N S -12 colors are used to construct the color distribution histogram of L. The intersection data of the background image and the digital camouflage image is calculated. Then, the intersection data and the sum of the pixels of the image are proportionally calculated to obtain the color similarity as shown in Equation (7).

[0027]

[0028] In the formula, N L-t (j) represents the number of pixels of the j-th color in the color histogram of the digital camouflage image, N L-b (j) represents the number of pixels of the j-th color in the background image color histogram. j represents the color category, and the maximum number of categories is calculated to be 9N. H +3N V +N S -12 types.

[0029] Color similarity S C The value ranges from (0,1). The larger the value, the more similar the color features of the digital camouflage image and the background image are.

[0030] S33, the texture similarity S TIt is obtained through the following method:

[0031] Texture evaluation is based on the principle of independent contrast. The background image and the digital camouflage image are scaled to the same size and converted to grayscale. Then, each image is divided into sub-blocks of size m×m, and all sub-blocks are rearranged into a one-dimensional sequence from left to right and top to bottom. Subsequently, a first-order two-dimensional discrete wavelet transform is performed on all sub-blocks of both images to obtain cA1(k), cH1(k), cV1(k), and cD1(k) components. cA1(k) represents the first-order approximation coefficient of the k-th sub-block, cH1(k) represents the first-order horizontal detail coefficient of the k-th sub-block, cV1(k) represents the first-order vertical detail coefficient of the k-th sub-block, and cD1(k) represents the first-order diagonal detail coefficient of the k-th sub-block. Furthermore, cA1, cH1, cV1, and cD1 can represent the first-order approximation, horizontal, vertical, and diagonal coefficients of the entire image, respectively. Further, multi-order two-dimensional discrete wavelet transforms can be performed on the cA1 wavelet components of the two images as needed, using cA... n cH n cV n , cD n These represent the four components of the nth order of the image. The three first-order parameters of the kth sub-block are extracted as texture features according to equations (8), (9), and (10).

[0032]

[0033] C31(k)=|cA1| (10)

[0034] In the formula, C11(k) is the first-order spatial frequency ratio parameter, representing the ratio of the first-order high-frequency component to the first-order low-frequency component of the k-th sub-block, used to measure the spatial frequency of the texture; C21(k) is the first-order polarization direction ratio parameter, representing the ratio of the first-order horizontal component to the first-order vertical component of the k-th sub-block, used to measure the polarization direction of the texture; C31(k) is the first-order low-frequency energy parameter, representing the first-order low-frequency energy of the texture of the k-th sub-block, used as a measure of the contrast between blocks. The first-order C11, C21, and C31 parameters of the image are obtained by summing the components of all blocks of the image, as shown in equations (11), (12), and (13).

[0035]

[0036] If we perform multi-order two-dimensional discrete wavelet extraction on the background image and the digital camouflage image, we can obtain the feature sequence [C11,C12,……,C1 n ],[C21,C22,……,C2 n [C31,C32,……,C3] nThe texture similarity evaluation index is defined, and the specific calculation formula is shown in equations (14)(15)(16)(17).

[0037]

[0038] In the formula, d1, d2, and d3 represent the spatial frequency ratio difference, polarization direction ratio difference, and low-frequency energy difference between the digital camouflage image and the background image, respectively. These represent the spatial frequency ratio, polarization direction ratio, and low-frequency energy parameter of the nth-order digital camouflage image and the background image, respectively.

[0039] Texture similarity S T The output range is (0,1). The larger the value, the more similar the texture features of the digital camouflage image and the background image are.

[0040] S34, the similarity of patch size S P It is obtained through the following method:

[0041] The background image and the digital camouflage image are converted to grayscale images respectively. Multiple thresholds are then selected to convert the grayscale images to binary images. The number of connected regions in the current background binary image is counted as the number of patches. The area of ​​the smallest patch region in the background binary image is A. P-b-min Count the number of connected regions in a binary image of digital camouflage, where the smallest patch area is A. P-t-min Subsequently, morphological operations were performed on the background binary image and the digital camouflage binary image, in the following order: image morphological opening operation, image morphological closing operation, and the radius parameter R in the structuring element required for the intermediate operations. SE It is calculated from equation (18).

[0042]

[0043] The patch sizes in digital camouflage images are larger than the patch sizes in the background image, therefore choosing an appropriate structuring element radius is crucial. After morphological processing, the patch sizes of connected regions in the background binary image and the digital camouflage binary image are re-statistically analyzed to identify the area of ​​the largest patch in each image. Patch size similarity S P The calculation formula is shown in equation (19).

[0044]

[0045] In the formula, A P-b-max and A P-b-min Let A represent the maximum and minimum connected patch areas of the background binary image. P-t-max and A P-t-minThis represents the maximum and minimum connected patch areas of the binary image of digital camouflage. For camouflage effectiveness, the maximum patch area in the binary image of digital camouflage should not exceed that of the background image patches. The optimal ratio parameter is set to 0.5, and the output range for patch size similarity is (0,1). The larger the value, the more similar the patch size ratio features are between the digital camouflage image and the background image.

[0046] S35, the edge similarity S E It is obtained through the following method:

[0047] Edge similarity evaluation is based on the principle of fusion contrast, placing the digital camouflage image at the center of the background image. Then, according to requirements, edge pixels of the background image and the digital camouflage image are selected for intersection, and Gabor filtering is applied to calculate the parallel energy |E| of the edge pixels. p |and orthogonal energy|E o Parallel energy represents the degree of blurring of continuous edges, while orthogonal energy represents the degree of blurring of non-real edges. Furthermore, the edge blur ratio R is calculated for each edge pixel. E As shown in equation (20).

[0048]

[0049] To accurately assess edge similarity, multiple sets of angle and wavelet wavelength combinations were selected to calculate the comprehensive data R of the edge blur ratio. E-ALL The calculation results are shown in equation (21).

[0050]

[0051] In the formula, angle represents the angle offset calculated by Gabor filtering, and the alternative value is calculated as shown in formula (22). wavelength represents the wavelet wavelength parameter calculated by Gabor filtering, and the alternative value is calculated as shown in formula (23).

[0052]

[0053] In the formula, p and q both represent positive integers, angle takes a flexible value according to the complexity of the calculation, and wavelength represents the smaller value of the length and width of the pixels involved in the calculation.

[0054] Furthermore, based on the fuzzy ratio comprehensive data obtained from equation (21), the average fuzzy ratio of all edge pixels is calculated, thus obtaining the local information edge similarity S. E The calculation is shown in equation (24).

[0055]

[0056] In the formula, N E This represents the number of edge pixels. The edge similarity parameter is the ratio of false edges to true edge contours in a measured edge profile; this value ranges from 0 to 1. Edge similarity S E The higher the value, the higher the edge similarity between the background image and the digital camouflage image.

[0057] S37, the edge blending degree S H It is obtained through the following method:

[0058] Edge fusion evaluation is based on the principle of fusion contrast. The digital camouflage image is placed at the center of the background image, and then the fused image is converted into a grayscale image. Then, an appropriate threshold is selected according to actual needs to convert the grayscale image into a binary image. The binary region where the background binary image and the digital camouflage binary image meet is taken, and the width is determined according to actual needs. The principle of edge fusion evaluation is based on the degree of damage to the original background image's meeting edge by the meeting edge of the digital camouflage image. The difference in edge density of the images is used to measure the degree of damage to the meeting edge. The closer the edge density parameters of the background image and the digital camouflage image are, the better the camouflage effect is. Conversely, the greater the difference, the less damage to the meeting edge is caused, and the worse the camouflage effect is. The edge density calculation of the meeting region of the background image and the digital camouflage image is shown in equations (25) and (26).

[0059]

[0060] In the formula, N H-b A represents the total number of edges in the mating region of the background image. H-b This indicates the number of pixels in the background image docking area. N H-t A represents the total number of edges in the docking region of a digital camouflage image. H-t This indicates the number of pixels in the docking area of ​​a digital camouflage image. Edge blending degree S H The calculation method is shown in equation (27).

[0061]

[0062] Edge blending degree S H The calculation result ranges from (0,1). The larger the value, the better the camouflage edge fusion characteristics of the digital camouflage image and the background image.

[0063] S38. According to steps S31 to S37, evaluate the digital camouflage effect according to the comprehensive evaluation index described in formula (1) to obtain the digital camouflage evaluation result. The output range of the comprehensive evaluation index is (0,1). The larger the value, the better the camouflage performance.

[0064] Beneficial effects

[0065] 1. This invention discloses a method for evaluating the camouflage effect of digital camouflage, which constructs a gray-level similarity S based on gray-level histogram statistical data. G The model constructs a color similarity S based on a color distribution histogram in the HSV space. C The model is constructed based on the texture similarity S of images using two-dimensional discrete wavelet transform. T The model constructs a patch size similarity S based on the statistical distribution information of connected patch regions in an image. P Model, based on edge similarity S using Gobor filtering E The model, based on edge blending degree S of edge contour density information. H The model, by constructing models of grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge fusion degree, enables quantitative evaluation of the above camouflage indicators. The quantitative analysis method improves the evaluation accuracy of digital camouflage, and has typicality and effectiveness. In anti-camouflage reconnaissance work, it can effectively detect target features and save computing resources.

[0066] 2. The digital camouflage effect evaluation method disclosed in this invention adopts a scheme of independent scoring of multiple features and then overall weighted fusion, which has transparency and objectivity. In the digital camouflage design work, the scoring of sub-evaluation parameters can provide effective guidance for the design and optimization of specified structures of digital camouflage.

[0067] 3. The digital camouflage effect evaluation method disclosed in this invention adds edge similarity and edge fusion degree local evaluation parameters to the traditional overall detection features, which include grayscale similarity, color similarity, texture similarity, and patch size similarity. It also compares the digital camouflage image with the background image, which is objective, reasonable and comprehensive. Attached Figure Description

[0068] Figure 1 A flowchart for evaluating the camouflage effect of digital camouflage;

[0069] Figure 2 Background image;

[0070] Figure 3 Image A is a digital camouflage image.

[0071] Figure 4 Image B is a digital camouflage image.

[0072] Figure 5 Image C is a digital camouflage image;

[0073] Figure 6 The background image is incorporated into the digital camouflage image C;

[0074] Figure 7 This is the grayscale distribution histogram of the background image;

[0075] Figure 8 This is the grayscale distribution histogram of digital camouflage image A;

[0076] Figure 9 The grayscale distribution histogram of digital camouflage image B;

[0077] Figure 10 Here is the grayscale distribution histogram of the digital camouflage image C;

[0078] Figure 11 This is a histogram of the color distribution of the background image.

[0079] Figure 12 The color distribution histogram of digital camouflage image A;

[0080] Figure 13 The color distribution histogram of digital camouflage image B;

[0081] Figure 14 The color distribution histogram of digital camouflage image C;

[0082] Figure 15 A binary image of the background image;

[0083] Figure 16 This is a binary image of digital camouflage image A;

[0084] Figure 17 This is a binary image of digital camouflage image B;

[0085] Figure 18 The binary image of digital camouflage image C;

[0086] Figure 19 A binary background image that undergoes morphological synchronous processing for both the background image and the digital camouflage image A;

[0087] Figure 20 A binary image of digital camouflage A, which is morphologically processed simultaneously for the background image and the digital camouflage image A.

[0088] Figure 21 A binary background image that undergoes morphological synchronous processing for both the background image and the digital camouflage image;

[0089] Figure 22 A binary image of digital camouflage B that undergoes morphological synchronous processing of the background image and the digital camouflage camouflage image B.

[0090] Figure 23The binary background image obtained by morphological synchronous processing of the background image and the digital camouflage image C;

[0091] Figure 24 The binary digital camouflage B image obtained by morphological synchronous processing of the background image and the digital camouflage image C;

[0092] Figure 25 The "return" - shaped docking edge area between the digital camouflage image and the background image;

[0093] Figure 26 The "return" - shaped binary image of the docking area between the digital camouflage image A and the background image;

[0094] Figure 27 The "return" - shaped binary image of the docking area between the digital camouflage image B and the background image;

[0095] Figure 28 The "return" - shaped binary image of the docking area between the digital camouflage image C and the background image;

[0096] Figure 29 The edge image of the docking area between the digital camouflage image A and the background image;

[0097] Figure 30 The edge image of the docking area between the digital camouflage image B and the background image;

[0098] Figure 31 The edge image of the docking area between the digital camouflage image C and the background image. Specific implementation method

[0099] To make the purpose, technical solution and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with specific embodiments and accompanying drawings.

[0100] As Figure 1 shown, a method for evaluating the digital camouflage effect disclosed in this embodiment is specifically implemented as follows:

[0101] Step 1: Use aerial photography equipment such as drones to photograph and record the background image. The computing platform reads the background image, as Figure 2 shown. Read three digital camouflage images A, B, and C with different parameter structures for comparative verification experiments, as Figure 3 , Figure 4 and Figure 5 shown. Place the digital camouflage image on top of the background image layer. By setting different length and width dimensions of the digital camouflage image, achieve the simulation effect of shooting and recording the reconnaissance target at different distances by aerial photography equipment such as drones. Use the geometric center of the digital camouflage image as the center point and expand the nine - grid area outward, as Figure 6 As shown, the actual background image pixel range used in the calculation is 900×900 pixels, and the digital camouflage image is 300×300 pixels.

[0102] Step 2: Select sub-evaluation parameters for evaluating the digital camouflage effect. The selected sub-evaluation parameters include: grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge blending degree. These parameters are also used as feature parameters for drones and other aerial photography equipment to detect camouflaged targets in the background.

[0103] Step 3: Calculate gray-level similarity parameters using the constructed gray-level similarity model, and then apply this to the background image. Figure 2 Digital camouflage images A, B, and C were scaled to 900×900 pixels and converted to grayscale images. Grayscale histogram data matrices were calculated, where the first row of the matrix contains grayscale levels ranging from 0 to 256, and the second row contains frequency data and the background image. Figure 2 The grayscale histograms of digital camouflage images A, B, and C are respectively as follows: Figure 6 , Figure 7 , Figure 8 and Figure 9 As shown. The grayscale similarity data of the background image and digital camouflage images A, B and C are obtained by using equation (1).

[0104]

[0105] In the formula, i represents the gray level, and N G-t (i) and N G-b (i) represent the number of pixels in the digital camouflage image and the background image at the i-th gray level, respectively.

[0106] The calculated grayscale similarity scores between digital camouflage images A, B, and C and background image 2 are 0.8581, 0.9070, and 0.9444, respectively. The blurriness levels of the three digital camouflage images, from highest to lowest, are A, B, and C. Image C has the lowest quantization level and a relatively high degree of restoration of the original background image; therefore, the three sets of data are consistent with the expected results.

[0107] Step 4: Calculate the color similarity parameters using the constructed color similarity model. Scale the background image and digital camouflage images A, B, and C to 900×900 pixels and convert them to HSV color space. Considering factors such as computational speed and storage capacity, quantize the color data in H, S, and V spaces according to equations (2), (3), and (4), respectively.

[0108]

[0109] In equation (2), h represents the hue value before quantization, and H represents the hue value after quantization. In equation (3), v represents the saturation value before quantization, and V represents the saturation value after quantization. In equation (4), s represents the lightness value before quantization, and S represents the lightness value after quantization.

[0110] Then, calculate the one-dimensional color vector L according to equation (5).

[0111] L=9H+3S+V (5)

[0112] Calculations show a total of 72 colors. This yields the color distribution histogram matrix data and color distribution histograms for the background image and digital camouflage images A, B, and C, respectively. The first row of the matrix represents the color level, and the second row represents the frequency data. The color distribution histograms are shown below. Figure 11 , Figure 12 , Figure 13 and Figure 14 As shown.

[0113] Furthermore, based on the color similarity evaluation index calculation formula (6), the color similarity parameters of the three sets of digital camouflage images were obtained respectively.

[0114]

[0115] In the formula, N L-t (j) represents the number of pixels of the j-th color in the color histogram of the digital camouflage image, N L-b (j) represents the number of pixels of the j-th color in the background image color histogram, where j represents the color category.

[0116] The calculated color similarity scores between digital camouflage images A, B, and C and the background image were 0.8388, 0.8456, and 0.8639, respectively. The three sets of results showed little difference and all had high scores, indicating that color similarity is not a key feature distinguishing the camouflage effectiveness of the three sets of digital camouflage images.

[0117] Step 5: Calculate the texture similarity parameters using the constructed texture similarity model. Scale the grayscale images corresponding to the background image and the three sets of digital camouflage images to the same size of 900×900, then divide them into sub-blocks of 10×10 pixels each. The digital camouflage images and the background image are divided into 90×90 sub-blocks. Rearrange all sub-blocks of the digital camouflage images A, B, C and the background image into a one-dimensional sequence according to the index order from left to right and from top to bottom. Then perform a two-dimensional Haar discrete wavelet transform on each sub-block, generating four components: cA1(k), cH1(k), cV1(k), and cD1(k) for each sub-block. Calculate the first-order spatial frequency ratio parameter C11, the first-order polarization direction ratio parameter C21, and the first-order low-frequency energy parameter C31 of the texture according to formulas (7), (8), and (9).

[0118]

[0119] The calculation results of C11, C21 and C31 of the background image and digital camouflage images A, B and C are shown in Table 1.

[0120] Table 1 Calculation Results for C11, C21, and C31

[0121]

[0122]

[0123] Furthermore, the larger C11, C21, and C31 values ​​in the background image and the digital camouflage image are compared and selected as the denominators. The ratio of the smaller value to the larger value constitutes the energy component evaluation indices d1, d2, and d3. The energy component evaluation indices are calculated as shown in equations (10), (11), and (12). Then, the texture similarity evaluation index is calculated according to equation (13).

[0124]

[0125] The calculated texture similarity scores between digital camouflage images A, B, and C and the background image are 0.1833, 0.3717, and 0.6558, respectively. The three sets of data differ significantly. As stated in step 3 regarding the grayscale similarity calculation, digital camouflage image A exhibits the highest degree of blurring and a low degree of restoration of various features of the original background, particularly evident in the texture similarity evaluation. Therefore, design optimization work is needed for the texture features of digital camouflage images A and B.

[0126] Step 6: Calculate the patch size similarity parameters using the constructed patch size similarity model, and convert the grayscale images of the background image and digital camouflage images A, B, and C into binary images, as shown below. Figure 15 , Figure 16 , Figure 17 and Figure 18 As shown. Then, preliminary statistics of connected region patch information were performed on the background image and the three sets of digital camouflage binary images, respectively, to obtain the minimum patch size area information for the background binary image and the digital camouflage binary image, which are A. P-b-min and A P-t-min The radius of the structuring element used for image morphology calculations is calculated according to formula (14).

[0127]

[0128] The calculated structuring element radii for the background binary image and the three sets of digital camouflage binary images are 2.7839, 2.4495, and 0.8660, respectively. The actual radii used in the calculation need to be rounded down, resulting in 3, 3, and 1, respectively. Subsequently, morphological opening operations were performed on the background binary image and the three sets of digital camouflage binary images to obtain three sets of morphologically processed binary images, as shown below. Figure 19 , Figure 20 , Figure 21 , Figure 22 , Figure 23 and Figure 24 As shown. Then, the maximum area data of connected patches in the background binary image and the digital camouflage binary images A, B, and C are calculated respectively. A P-b-max and A P-t-max The results are shown in Table 2.

[0129] Table 2. Statistics on the Maximum Area of ​​Connected Regions

[0130]

[0131] The patch size similarity parameter is calculated according to formula (15).

[0132]

[0133] The calculated patch size similarity results between digital camouflage images A, B, and C and the background image are 0.5832, 0.6388, and 0.8003, respectively. As stated in step 3, digital camouflage image A has the highest degree of blurring, therefore the patch size after morphological processing is also larger, and the difference in patch size between it and the background image is also greater. Therefore, it can be concluded that the patch size features of digital camouflage images A and B need to be improved and optimized to achieve a better fit with the background image.

[0134] Step 7: Calculate the edge similarity parameters using the constructed edge similarity model. Embed the digital camouflage images A, B, and C into the center positions of the background images respectively. Convert the fused images into grayscale images, and extract 20 rows of pixels at the docking edges between the digital camouflage images and the background images as the docking edge images of the two images. Among them, 10 rows of pixels are taken for both the background image and the original camouflage image to obtain a "hui"-shaped docking edge image, as Figure 25 shown.

[0135] Furthermore, perform Gabor filtering calculations on the two groups of edge images respectively. The combinations of wavelength parameters, horizontal angles, and vertical angles are shown in Table 3.

[0136] Table 3 Parameter combination table

[0137]

[0138]

[0139] Under each group of parameter combinations, calculate the horizontal / orthogonal energy ratios of the edges of the three groups of digital camouflage images and the edges of the background images, and the edge similarity parameter S respectively according to formulas (16), (17), and (18) E .

[0140]

[0141] In formula (14), |E o | and |E p | respectively represent the horizontal energy and orthogonal energy of the edge image, R E represents the horizontal / orthogonal energy ratio under a certain wavelength parameter and angle parameter, R E represents the sum of the horizontal / orthogonal energy ratios under all wavelength parameters and angle parameters, and N E represents the number of edge pixels.

[0142] Summarize the calculation results under multiple groups of parameter combinations as shown in Table 4.

[0143] Table 4 Summary of multi-parameter calculation results

[0144]

[0145] The calculated edge similarities of the three groups of digital camouflage images to the background images are 0.8263, 0.8584, and 0.8420 respectively. The data shows that the docking edges of the three groups of digital camouflage images and the background images all have good similarity and the differences are small.

[0146] Step 8: Calculate the edge fusion degree parameters using the constructed edge fusion degree model. First, embed the digital camouflage binary image into the background image. Based on the docking edge, select an appropriate width parameter. For the embedded digital camouflage binary image, expand the pixel area by a specified width inward. For the background image, expand the background image by a specified width outward to form a "hui" - shaped image. The fusion binary images of the three groups of digital camouflage images and the background image are shown in Figures (26), (27), and (28) respectively. Perform a morphological closing operation on the "hui" - shaped image area, and then use the Canny operator to extract its edges. The three groups of fused edge images are shown in Figures (29), (30), and (31) respectively. Calculate the edge fusion degree parameters of the three groups using formulas (19), (20), and (21).

[0147]

[0148] In the formula, N H-b and N H-t represent the number of edges of the background image and the digital camouflage image respectively, A H-b and A H-t represent the number of pixels of the background image and the digital camouflage image respectively, ρ e-b and ρ e-t represent the edge density of the background image and the digital camouflage image respectively.

[0149] The calculation results of the edge fusion degree parameters of the three groups are: 0.0663, 0.1846, and 0.7152 respectively. The results show that there are significant differences in the edge fusion characteristics between the three groups of digital camouflage images and the background image. It can be concluded that there is a large room for improvement in the edge density characteristic information of digital camouflage images A and B.

[0150] Step 9: Perform parameter fusion on the gray - scale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge fusion degree parameters obtained in the above steps according to the weights obtained by the previous experimental experience method to obtain the final similarity evaluation index. The calculation formula is shown in Equation (22).

[0151] S = a×S G +b×S C +c×S T +d×S P +e×S E +f×S H (22)

[0152] The summary data of the similarity parameters of the three groups of digital camouflage images and the background image are shown in Table 5.

[0153] Table 5 Summary Table of Similarity Parameters

[0154]

[0155] In the formula, the weight of grayscale similarity parameter a is 0.23, the weight of color similarity parameter b is 0.21, the weight of texture similarity parameter c is 0.10, the weight of patch size similarity is 0.17, the weight of edge similarity parameter e is 0.21, and the weight of edge fusion parameter f is 0.08. The comprehensive evaluation results of the three sets of digital camouflage images in the background image are calculated to be 0.6696, 0.7270, and 0.8343, respectively.

[0156] The method also includes step 10, which guides and optimizes the design of digital camouflage based on the specific data obtained in step 9. The results of three sets of comparative experiments show that digital camouflage image C has the best camouflage effect, followed by image B, with image A ranking last. The evaluation results are consistent with the actual camouflage effects of the three camouflage images. Specifically, the edge blending features, texture features, and patch size features of digital camouflage images A and B differ significantly from the background image. Therefore, targeted optimization and correction can be made to the design of digital camouflage images to improve the camouflage performance of the target in various scenarios. The digital camouflage evaluation method involved in this embodiment can effectively evaluate the camouflage effect of digital camouflage. Furthermore, the various sub-evaluation parameters can clearly identify the defects of digital camouflage images. In the analysis of this embodiment, the data shows that the texture design and edge blending design of digital camouflage images A and B are relatively poor.

[0157] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. 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 within the scope of protection of the present invention.

Claims

1. A method for evaluating the camouflage effect of digital camouflage, characterized in that: Includes the following steps: S1. Use aerial photography equipment to collect background images of the actual work scene, and at the same time obtain digital camouflage images that match the background; S2. Use the computing platform to obtain the size of the digital camouflage image; determine the target camouflage position coordinates of the digital camouflage image in the original background image; use these position coordinates as the center to divide the original background image into a nine-grid area, with the length and width of each grid cell being consistent with the size of the camouflage image; this nine-grid area is the background image that actually participates in the feature extraction calculation. S3. Store the processed background image and digital camouflage image into the computing platform. The computing platform contains multiple evaluation parameter models for assessing the digital camouflage effect, specifically including the grayscale similarity S between the background image and the digital camouflage image. G Color similarity S C Texture similarity S T Patch size similarity S P Edge similarity S E Edge blending degree S H The above indicators are combined according to certain weighting parameters: S=a×S G +b×S C +c×S T +d×S P +e×S E +f×S H (1) In the formula, a, b, c, d, e, and f represent the weight parameters of grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge fusion degree, respectively. The digital camouflage effect is evaluated based on the comprehensive evaluation index, and the evaluation result of digital camouflage is obtained. The output range of the comprehensive evaluation index is (0,1). The larger the value, the better the camouflage performance. The specific implementation method of S3 is as follows: S30. Select evaluation parameters for assessing the digital camouflage effect. The evaluation parameters include the grayscale similarity S between the background image and the digital camouflage image. G Color similarity S C Texture similarity S T Patch size similarity S P Edge similarity S E Edge blending degree S H ; S31, the grayscale similarity S G We obtain it from the following formula: Gray-scale evaluation is based on the principle of independent comparison. The background image and the digital camouflage image are scaled to the same size and converted into gray-scale images. Then, the two images are scaled to the same size, and their respective gray-scale histograms are constructed. The intersection data of the two gray-scale histograms is calculated. Then, the intersection data of all gray-scale levels is proportionally calculated with the sum of the pixels of the image to obtain the gray-scale similarity as shown in Equation (2). In the formula, N G-t (i) represents the number of pixels with the i-th gray level value in the gray-level histogram of the digital camouflage image, N G-b (i) represents the number of pixels with the i-th gray value in the gray-level histogram of the background image; i represents the grayscale level, with a maximum output of 256 levels; Gray-scale similarity S G The output range is (0,1). The larger the value, the more similar the grayscale features of the digital camouflage image and the background image are. S32, the color similarity S C It is obtained through the following method: Color evaluation is based on the principle of independent contrast. The background image and the digital camouflage image are scaled to the same size and converted to HSV space. The colors are further quantized according to actual needs. The quantization formulas are shown in equations (3)(4)(5). In equation (3), h represents the hue value before quantization, H represents the hue value after quantization, and N... H -1 is the maximum output value of the quantized hue, C p The interval bias value of hue before quantization is given; in equation (4), v represents the saturation value before quantization, V represents the saturation value after quantization, and N represents the saturation value after quantization. V -1 represents the maximum output value of saturation after quantization, k1, k2, ... It is the interval boundary value of saturation before quantization; in equation (5), s represents the brightness value before quantization, S represents the brightness value after quantization, and N S -1 represents the maximum output value of the quantized brightness, k1, k2, ... These are the boundary values ​​of each interval; where the quantization interval boundary values ​​of V and S are consistent; the color components are synthesized into a one-dimensional vector according to equation (6); L=9H+3S+V (6) The color range L becomes 9N. H +3N V +N S -12 colors, construct the color distribution histogram of L, calculate the intersection data of the background image and the digital camouflage image, and then perform a proportional operation on the intersection data and the total number of pixels of the image to obtain the color similarity as shown in Equation (7). In the formula, N L-t (j) represents the number of pixels of the j-th color in the color histogram of the digital camouflage image, N L-b (j) represents the number of pixels of the j-th color in the background image color histogram; j represents the color category, and the maximum number of categories calculated is 9N. H +3N V +N S -12 types; Color similarity S C The value ranges from (0,1). The larger the value, the more similar the color features of the digital camouflage image and the background image are. S33, the texture similarity S T It is obtained through the following method: Texture evaluation is based on the independent contrast principle. The background image and the digital camouflage image are scaled to the same size and converted to grayscale. Then, each image is divided into m×m sub-blocks, and all sub-blocks are rearranged into a one-dimensional sequence from left to right and top to bottom. Subsequently, a first-order two-dimensional discrete wavelet transform is performed on all sub-blocks of both images to obtain cA1(k), cH1(k), cV1(k), and cD1(k) components. cA1(k) represents the first-order approximation coefficient of the k-th sub-block, cH1(k) represents the first-order horizontal detail coefficient of the k-th sub-block, cV1(k) represents the first-order vertical detail coefficient of the k-th sub-block, and cD1(k) represents the first-order diagonal detail coefficient of the k-th sub-block. Furthermore, cA1, cH1, cV1, and cD1 represent the first-order approximation, horizontal, vertical, and diagonal coefficients of the entire image, respectively. Further, multi-order two-dimensional discrete wavelet transforms are performed on the cA1 wavelet components of the two images as needed, using cA... n cH n cV n , cD n These represent the four components of the nth order of the image; according to equations (8), (9), and (10), the three first-order parameters of the kth sub-block are extracted as texture features; C31(k)=|cA1| (10) In the formula, C11(k) is the first-order spatial frequency ratio parameter, which represents the ratio of the first-order high-frequency component to the first-order low-frequency component of the k-th sub-block, and is used to measure the spatial frequency of the texture; C21(k) is the first-order polarization direction ratio parameter, which represents the ratio of the first-order horizontal component to the first-order vertical component of the k-th sub-block, and is used to measure the polarization direction of the texture; C31(k) is the first-order low-frequency energy parameter, which represents the first-order low-frequency energy of the texture of the k-th sub-block, and is used as the basis for measuring the contrast between blocks; the first-order C11, C21, and C31 parameters of the image are obtained by summing the components of all blocks of the image, and the calculation is shown in formulas (11)(12)(13). If multi-order two-dimensional discrete wavelet extraction is performed on the background image and the digital camouflage image, the feature sequence [C11,C12,……,C1] is obtained. n ],[C21,C22,……,C2 n [C31,C32,……,C3] n Define texture similarity evaluation index, and the specific calculation formula is shown in equations (14)(15)(16)(17). In the formula, d1, d2, and d3 represent the spatial frequency ratio difference, polarization direction ratio difference, and low-frequency energy difference between the digital camouflage image and the background image, respectively. These represent the spatial frequency ratio, polarization direction ratio, and low-frequency energy parameter of the nth-order digital camouflage image and the background image, respectively. Texture similarity S T The output range is (0,1). The larger the value, the more similar the texture features of the digital camouflage image and the background image are. S34, the similarity of patch size S P It is obtained through the following method: The background image and digital camouflage image are converted to grayscale images, and multiple thresholds are selected to convert the grayscale images to binary images. The number of connected regions in the current background binary image is counted as the number of patches, and the area of ​​the smallest patch region in the background binary image is A. P-b-min Count the number of connected regions in a binary image of digital camouflage, where the smallest patch area is A. P-t-min Subsequently, morphological operations were performed on the background binary image and the digital camouflage binary image, in the following order: image morphological opening operation, image morphological closing operation, and the radius parameter R in the structuring element required for the intermediate operations. SE Calculated by equation (18); The patch size in digital camouflage images is larger than the patch size in the background image, therefore choosing an appropriate structuring element radius is crucial. After morphological processing, the patch size data of connected regions in the background binary image and the digital camouflage binary image are re-statistically analyzed to find the area of ​​the largest patch in each image. The patch size similarity S... P The calculation formula is shown in equation (19); In the formula, A P-b-max and A P-b-min Let A represent the maximum and minimum connected patch areas of the background binary image. P-t-max and A P-t-min This represents the maximum and minimum connected patch areas of the digital camouflage binary image. In terms of camouflage effect, the maximum patch area in the digital camouflage binary image must not exceed the patch area in the background image. The optimal ratio parameter is set to 0.5, and the output range of patch size similarity is (0,1). The larger the value, the more similar the patch size ratio features of the digital camouflage image and the background image are. S35, the edge similarity S E It is obtained through the following method: Edge similarity evaluation is based on the principle of fusion contrast, placing the digital camouflage image at the center of the background image; then, according to requirements, edge pixels of the background image and digital camouflage image are selected for docking, and Gabor filtering is applied to calculate the parallel energy |E| of the edge pixels. p |and orthogonal energy|E o Parallel energy represents the degree of blurring of continuous edges, while orthogonal energy represents the degree of blurring of non-real edges; furthermore, the edge blur ratio R is calculated for each edge pixel. E As shown in equation (20); To accurately assess edge similarity, multiple sets of angle and wavelet wavelength combinations were selected to calculate the comprehensive data R of the edge blur ratio. E-ALL The calculation results are shown in equation (21); In the formula, angle represents the angle offset calculated by Gabor filtering, and the alternative values ​​are calculated as shown in formula (22); wavelength represents the wavelet wavelength parameter calculated by Gabor filtering, and the alternative values ​​are calculated as shown in formula (23). In the formula, p and q both represent positive integers, angle takes a flexible value according to the complexity of the calculation, and wavelength represents the smaller value of the length and width of the pixels involved in the calculation. Furthermore, based on the fuzzy ratio comprehensive data obtained from equation (21), the average fuzzy ratio of all edge pixels is calculated, thus obtaining the local information edge similarity S. E The calculation is shown in equation (24); In the formula, N E Indicates the number of pixels at the edge; The edge similarity parameter is the ratio of the false edge to the true edge contour in a measurement of the edge profile. This value ranges from 0 to 1. The edge similarity S... E The higher the value, the higher the edge similarity between the background image and the digital camouflage image; S37, the edge blending degree S H It is obtained through the following method: The edge fusion evaluation is based on the principle of fusion contrast. The digital camouflage image is placed at the center of the background image. Then, the fused image is converted into a grayscale image. Then, according to the actual needs, an appropriate threshold is selected to convert the grayscale image into a binary image. The binary region where the background binary image and the digital camouflage binary image meet is taken, and the width is determined according to the actual needs. The edge fusion evaluation principle is based on the degree of damage of the connecting edge of the digital camouflage image to the connecting edge of the original background image. The edge density difference of the image is used to measure the degree of damage to the connecting edge. The closer the edge density parameters of the background image and the digital camouflage image are, the better the camouflage effect is. Conversely, the greater the difference, the less damage to the connecting edge is, and the worse the camouflage effect is. The edge density calculation of the connecting region of the background image and the digital camouflage image is shown in Equations (25) and (26). In the formula, N H-b A represents the total number of edges in the mating region of the background image. H-b N represents the number of pixels in the mating area of ​​the background image; H-t A represents the total number of edges in the docking region of a digital camouflage image. H-t This indicates the number of pixels in the docking area of ​​a digital camouflage image; edge blending degree S H The calculation method is shown in equation (27); Edge blending degree S H The calculation result ranges from (0,1). The larger the value, the better the camouflage edge fusion characteristics of the digital camouflage image and the background image. S38. According to steps S31 to S37, evaluate the digital camouflage effect according to the comprehensive evaluation index described in formula (1) to obtain the digital camouflage evaluation result. The output range of the comprehensive evaluation index is (0,1). The larger the value, the better the camouflage performance.

2. The method for evaluating the camouflage effect of digital camouflage as described in claim 1, characterized in that: The process also includes step S4: Based on the digital camouflage evaluation results obtained in step S3, multiple feature information, including grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge fusion degree, is independently extracted from the background image and the digital camouflage image; a comprehensive evaluation index that fuses multiple feature information is used to effectively evaluate the camouflage effect of the digital camouflage image in the background image; and objective guidance is provided for the design of digital camouflage based on the evaluation data of each sub-parameter of grayscale similarity, color similarity, texture similarity, patch size similarity, edge similarity, and edge fusion degree, thereby solving related engineering problems.

3. The method for evaluating the camouflage effect of digital camouflage as described in claim 1, characterized in that: The aerial photography equipment includes drones, satellites, unmanned airships, helicopters, airplanes, and parachutes.

4. The method for evaluating the camouflage effect of digital camouflage as described in claim 1, characterized in that: The computing platform includes an x86 CPU platform, an ARM SOC platform, an NVIDIA GPU, and its Jetson edge computing processor.

Images