Deformation camouflage spot generation method based on superpixel segmentation and fuzzy c-means clustering

Abstract

The application discloses a kind of based on superpixel segmentation and fuzzy C means clustering deformation dappled spot generation method, specifically is: first, obtain background picture, cut to same proportion with target;Then calculate the number of superpixel segmentation, the superpixel segmentation is carried out to background picture, other colors in superpixel are replaced by average color, and the color clustering of superpixel is carried out, using dappled main color replaces the color in chart, obtains big spot deformation dappled pattern;Edge extraction is carried out to the picture after clustering, obtains the spot pattern of big spot deformation dappled, extracts the shape of big spot of same color, generates small spot in each big spot using the algorithm of superpixel segmentation and fuzzy C means clustering combination;Finally, the small spot pattern obtained by each big spot pattern is superimposed, and the spot pattern of small spot deformation dappled is obtained.The application solves the connection and island problem of spot, improves the organic nature of the combination between irregular spots, meets the design requirements of deformation dappled.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital image processing, in particular to a deformed camouflage spot generation method based on superpixel segmentation and fuzzy C-means clustering. BACKGROUND

[0002] The deformed camouflage is a multi-color camouflage composed of irregular spots in shape and size. When the target is in different backgrounds, on the one hand, the irregular spots of the deformed camouflage distort the straight contour of the target, and on the other hand, the design of the spot shape and color produces a fusion or brightness highlighting effect with the background, so that the target produces a camouflage deformation effect, which is suitable for the camouflage protection of the mobile target in different backgrounds.

[0003] The design of the irregular shape of the deformed camouflage spot is drawn according to certain principles, which include asymmetry, no right angle, no long straight line, etc. With the development of computer technology, the design of the camouflage is combined with computer image processing technology. The computer language expects to obtain a set of accurate data to represent a certain determined spot. Some researchers represent the irregular spot shape in the form of a feature vector and establish a corresponding spot library. Establishing the irregular spot library indeed obtains many irregular spots, but the combination between the irregular spots becomes a difficult problem. For example, the contact between two convex spots cannot be perfectly connected, and the other boundaries outside the contact point will produce a blank. Some researchers use the fractal Brown model to directly generate a whole camouflage pattern to solve the combination problem between the spots, but the Brown fractal model is suitable for simulating high-resolution terrain. The flat ground is simulated as a concave and convex ground, and the intersection line obtained by cutting the terrain with the contour surface will form some irregular spots, but the sizes of these spots are not the same, which cannot fully meet the size requirements of the spots. SUMMARY

[0004] The purpose of the present application is to provide a deformed camouflage spot generation method based on superpixel segmentation and fuzzy C-means clustering, so as to solve the connection and island problems of irregular spots and improve the organic combination between irregular spots to meet the design requirements of the deformed camouflage.

[0005] The technical solution for achieving the purpose of the present application is as follows: a deformed camouflage spot generation method based on superpixel segmentation and fuzzy C-means clustering, comprising the following steps:

[0006] Step 1, obtaining a background picture of a natural background;

[0007] Step 2, cutting the background picture into the same length-width ratio as the target to be camouflaged;

[0008] Step 3, calculating the number K of superpixel segmentation according to the area of the target and the reconnaissance resolution of the reconnaissance equipment, and performing superpixel segmentation on the background picture;

[0009] Step 4: Within each superpixel, replace the other colors with the average color to obtain a superpixel pattern with the average color.

[0010] Step 5: Cluster the colors of the superpixels into M colors, replace the colors of the clustered superpixels with the colors of the cluster centers to generate large spot deformed camouflage patterns of M colors, and replace the M colors in the image with the camouflage main color selected as needed to obtain the corresponding large spot deformed camouflage patterns.

[0011] Step 6: Extract edges from the clustered images to obtain the spotted pattern of the large spotted deformed camouflage.

[0012] Step 7: Assign a unique label to each color patch. All pixels within the same color patch have the same label. Use the labels to extract the shapes of large patches of the same color to obtain the background image of the large patch shapes for each color.

[0013] Step 8: Use an algorithm that combines superpixel segmentation and fuzzy C-means clustering to generate small spots within each large spot;

[0014] Step 9: Overlay the small spot patterns obtained from each large spot to obtain a small spot deformed camouflage pattern.

[0015] Furthermore, the step 1 of obtaining a background image with a natural background is as follows:

[0016] A background image of a natural background is acquired using an image acquisition device. The image contains only natural scenery and no man-made objects.

[0017] Furthermore, step 2, which involves cropping the background image to the same aspect ratio as the target to be disguised, is detailed as follows:

[0018] Measure the size of the target to be disguised, calculate the target's aspect ratio, and crop the background image to match the target's aspect ratio. The cropped background image contains N pixels.

[0019] Furthermore, in step 3, the number K of superpixel segments is calculated based on the target's area and the reconnaissance resolution of the reconnaissance equipment, and the background image is then segmented into superpixels, as detailed below:

[0020] The SLIC superpixel segmentation algorithm is used to divide the cropped background image into K superpixels of the same size. The size of the superpixel is S = sqrt(N / K), and the size of the spots in the morphed camouflage pattern is... D is the observation distance;

[0021] To ensure that there is no spatial color mixing between the spots of the metamorphic camouflage, the minimum value of the spot size d of the metamorphic camouflage is equal to the value of the reconnaissance resolution, the area A of the target to be camouflaged is determined, and the number K of superpixel partitions is calculated 2 ;

[0022] The spot size d of the metamorphic camouflage is organically combined with the number K of superpixel partitions, and K is adjusted so that the designed spot pattern meets the size requirements of the metamorphic camouflage spots.

[0023] Further, the value of the number M of color cluster categories in step 5 is M = 3.

[0024] Further, step 7 assigns a unique label to each color patch, and the labels of all pixel points in the same color spot are the same. The label is used to extract the shape of the large spot of the same color, and the background picture of the large spot of each color is obtained, as follows:

[0025] Step 7.1, for each color patch, create a matrix with the same size as the original image as a mask;

[0026] Step 7.2, find the position of the color patch to be left in the mask matrix, and set its value to 1, and the rest to 0;

[0027] Step 7.3, multiply the background image with the mask, when the value in the mask matrix is 1, the background image does not change; when the value in the mask matrix is 0, the background image becomes black, thereby extracting the background image of the corresponding shape.

[0028] Further, the algorithm of step 8 combines superpixel segmentation and fuzzy C-means clustering to generate small spots in each large spot, as follows:

[0029] Step 8.1, according to the ratio of the camouflage unit area of the small spot to the camouflage unit of the large spot, calculate the number K' of superpixel partitions of the small spot, and perform superpixel segmentation on the background picture;

[0030] Step 8.2, replace other colors with average color inside each superpixel of the small spot to obtain the superpixel pattern of the average color of the small spot;

[0031] Step 8.3, cluster the colors of the superpixels of the small spot into M' colors, replace the colors of the superpixels of the same cluster with the color of the cluster center, generate M' color small spot metamorphic camouflage patterns, and replace the M' colors in the pattern with the selected main color of the camouflage as needed to obtain the corresponding small spot metamorphic camouflage.

[0032] Further, the value of the number K' of the superpixel segmentation of the small spots in step 8.1 is 4 times the number K of the superpixel segmentation of the large spots, that is, K' = 4K.

[0033] Further, the value of the number M' of the color categories of the superpixels of the small spots in step 8.3 is M' = 2.

[0034] Further, the small spot pattern of the deformed camouflage is obtained by superimposing the small spot patterns obtained by each large spot, and the specific process is as follows:

[0035] After the small spot pattern of the deformed camouflage is generated, the colors in the two-color clustering map are replaced by the main color of the camouflage, and different spots are superimposed to generate a small spot deformed camouflage pattern.

[0036] Compared with the prior art, the present application has the following advantages:

[0037] (1) The spot pattern of the small spot deformed camouflage designed in the present application organically combines the K of the superpixel and the size of the spot, and follows the spot size principle of the deformed camouflage;

[0038] (2) The single spot meets the irregular principle of the spot, and the whole camouflage spot pattern is generated from the background as a whole, which solves the problems of spot connection and island shape, improves the organic combination between irregular spots, and meets the design requirements of the deformed camouflage. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is the flowchart of the deformed camouflage spot generation method based on superpixel segmentation and fuzzy C-means clustering of the present application.

[0040] Figure 2 is the grassland background map used for generating the camouflage spot pattern in the embodiment of the present application.

[0041] Figure 3 is the deformed camouflage pattern of the large spot generated by the algorithm combining superpixel segmentation and fuzzy C-means clustering in the embodiment of the present application, wherein (a) is the original map, (b) is the spot boundary map generated by superpixel segmentation, (c) is the three-color clustering map generated by fuzzy C-means, and (d) is the deformed camouflage spot map obtained by edge extraction on the clustered picture.

[0042] Figure 4 is the deformed camouflage pattern of the large spot obtained by replacing the colors in the map with the main color of the camouflage in the embodiment of the present application.

[0043] Figure 5are the background images of large spots of each color obtained by extracting large spot shapes of the same color in the embodiments of the present application, wherein (a) is a grassland background image for generating a deformed camouflage spot, (b) is a grassland background image extracted by a mask corresponding to the main color 1, (c) is a grassland background image extracted by a mask corresponding to the main color 2, and (d) is a grassland background image extracted by a mask corresponding to the main color 3.

[0044] Figure 6 are deformed camouflage images of small spots in the large spot of the main color 1 generated by an algorithm combining superpixel segmentation and fuzzy C-means clustering in the embodiments of the present application, wherein (a) is a spot shape background image of the main color 1, (b) is a superpixel segmentation image of the main color 1, (c) is a two-color clustering image of the main color 1, and (d) is a small spot image of the main color 1.

[0045] Figure 7 are deformed camouflage images of small spots in the large spot of the main color 2 generated by an algorithm combining superpixel segmentation and fuzzy C-means clustering in the embodiments of the present application, wherein (a) is a spot shape background image of the main color 2, (b) is a superpixel segmentation image of the main color 2, (c) is a two-color clustering image of the main color 2, and (d) is a small spot image of the main color 2.

[0046] Figure 8 are deformed camouflage images of small spots in the large spot of the main color 3 generated by an algorithm combining superpixel segmentation and fuzzy C-means clustering in the embodiments of the present application, wherein (a) is a spot shape background image of the main color 3, (b) is a superpixel segmentation image of the main color 3, (c) is a two-color clustering image of the main color 3, and (d) is a small spot image of the main color 3.

[0047] Figure 9 are spot images of small spot deformed camouflage obtained by replacing the colors in the two-color clustering image of small spots with the main colors in the embodiments of the present application, wherein (a) is a large spot camouflage image of the main color 1, (b) is a small spot camouflage image of the main color 1, (c) is a large spot camouflage image of the main color 2, (d) is a small spot camouflage image of the main color 2, (e) is a large spot camouflage image of the main color 3, and (f) is a small spot camouflage image of the main color 3.

[0048] Figure 10 is a small spot deformed camouflage spot image obtained by superimposing the small spot patterns obtained from each large spot pattern in the embodiments of the present application. DETAILED DESCRIPTION

[0049] In view of the design problem of the deformed camouflage spot pattern, the superpixel segmentation and fuzzy C-means clustering are combined, the deformed camouflage is generated as a whole, the problem of combining spots to generate the deformed camouflage from the spot library is avoided, and the obtained camouflage pattern meets the shape and size requirements of the deformed camouflage spots. The deformed camouflage pattern generated by the application meets the irregular principle of the single spot, and the camouflage at the target edge is composed of different color spots, so that the shape of the target can be distorted and changed.

[0050] The application will be further described in detail below with reference to the drawings and specific embodiments.

[0051] In combination Figure 1 The application is a deformed camouflage spot generation method based on superpixel segmentation and fuzzy C-means clustering, comprising the following steps:

[0052] Step 1, obtaining a background picture of a natural background;

[0053] Step 2, cutting the background picture into the same length-width ratio as the target to be camouflaged;

[0054] Step 3, calculating the number K of superpixel segmentation according to the area of the target and the reconnaissance resolution of the reconnaissance equipment, and performing superpixel segmentation on the background picture;

[0055] Step 4, replacing other colors with average colors in each superpixel to obtain an average color superpixel pattern;

[0056] Step 5, clustering the colors of the superpixels into M colors, replacing the colors of the clustered superpixels with the colors of the cluster centers, generating a large spot deformed camouflage pattern of M colors, and replacing the M colors in the pattern with a selected main color of the camouflage to obtain a corresponding large spot deformed camouflage pattern;

[0057] Step 6, performing edge extraction on the clustered picture to obtain a spot pattern of the large spot deformed camouflage;

[0058] Step 7, assigning a unique label to each color spot, the labels of all pixel points in the same color spot being the same, and using the labels to extract the shapes of the large spots of the same color to obtain a background picture of the shape of the large spot of each color;

[0059] Step 8, using an algorithm combining superpixel segmentation and fuzzy C-means clustering to generate small spots in each large spot;

[0060] Step 9, superimposing the small spot patterns obtained from each large spot to obtain a small spot deformed camouflage pattern.

[0061] As a specific example, the background picture of the natural background obtained in step 1 is as follows:

[0062] A background picture of a natural background is acquired by an image acquisition device, and the picture only contains natural scenery and does not contain artificial targets.

[0063] As a specific example, the background picture is cropped to have the same length-width ratio as the target to be camouflaged in step 2, and the cropping is specifically as follows:

[0064] The size of the target to be camouflaged is measured, the length-width ratio of the target is calculated, and the background picture is cropped to have the corresponding target ratio. The cropped background picture contains N pixels.

[0065] As a specific example, the number K of superpixel segmentation is calculated according to the area of the target and the reconnaissance resolution of the reconnaissance device in step 3, and the background picture is superpixel segmented, and the superpixel segmentation is specifically as follows:

[0066] The cropped background picture is segmented into K superpixels of the same size by using the SLIC superpixel segmentation algorithm. The size S of the superpixel is sqrt(N / K), and the size of the spot of the metamorphic camouflage is D is the observation distance;

[0067] To ensure that there is no spatial color mixing between the spots of the metamorphic camouflage, the minimum value of the size d of the spot of the metamorphic camouflage is equal to the value of the reconnaissance resolution. The area A of the target to be camouflaged is determined, and the number K of superpixel segmentation is calculated as K=A / d 2 ;

[0068] The size d of the spot of the metamorphic camouflage is organically combined with the number K of superpixel segmentation. By adjusting K, the designed spot pattern meets the size requirement of the spot of the metamorphic camouflage.

[0069] The metamorphic camouflage needs to ensure that different spots are clearly visible under a certain reconnaissance resolution, so that a part of the spots are fused with the background, and a part of the spots are distinguished from the background, and the overall effect of changing the original shape of the target is achieved. Therefore, the size of the spot has a minimum value requirement, which is equal to the value of the reconnaissance resolution, which is to ensure that the spot of the metamorphic camouflage is visible under this reconnaissance condition. If the size of the spot is too small, the spot will be spatially mixed. Taking a small car as an example, if the spot is too small, the original red car may become a yellow car (from a distance, the camouflage is mixed into yellow), at which time the metamorphic effect is lost, and the shape is still the original shape of the car.

[0070] As a specific example, the value of the number M of color cluster categories in step 5 is M=3.

[0071] As a specific example, step 7 assigns a unique label to each color patch, and the labels of all pixels in the same color patch are the same. The label is used to extract the shape of the large patch of the same color, and the background image of each color is obtained. The specific steps are as follows:

[0072] Step 7.1, for each color patch, create a matrix with the same size as the original image as a mask;

[0073] Step 7.2, find the location of the color patch to be kept in the mask matrix, and set its value to 1, and the rest to 0;

[0074] Step 7.3, multiply the background image with the mask. When the value in the mask matrix is 1, the background image does not change; when the value in the mask matrix is 0, the background image becomes black, thereby extracting the corresponding shape background image.

[0075] As a specific example, step 8 uses the algorithm combining superpixel segmentation and fuzzy C-means clustering to generate small patches in each large patch. The specific steps are as follows:

[0076] Step 8.1, according to the ratio of the area of the small patch to the area of the large patch, calculate the number of superpixel segmentation K' of the small patch, and perform superpixel segmentation on the background image;

[0077] Step 8.2, replace other colors with average color inside each superpixel of the small patch to obtain the superpixel pattern of the average color of the small patch;

[0078] Step 8.3, cluster the colors of the superpixels of the small patch into M' colors, replace the colors of the superpixels of the same class with the color of the cluster center, generate a small patch deformation camouflage pattern of M' colors, and replace the M' colors in the pattern with the selected main color of the camouflage to obtain the corresponding small patch deformation camouflage.

[0079] As a specific example, the number of superpixel segmentation K' of the small patch in step 8.1 is 4 times the number of superpixel segmentation K of the large patch, i.e. K' = 4K.

[0080] As a specific example, the number of color types M' of the superpixels of the small patch in step 8.3 is M' = 2.

[0081] As a specific example, step 9 superimposes the small patch patterns obtained from each large patch to obtain a small patch deformation camouflage pattern. The specific steps are as follows:

[0082] After generating the small patch pattern of the deformation camouflage, replace the colors in the two-color clustering pattern with the main color of the camouflage, and superimpose different patches to generate a small patch deformation camouflage pattern.

[0083] The application will be further described in detail below in combination with the drawings and specific embodiments.

[0084] Embodiment

[0085] This embodiment takes a large equipment in a typical eastern background as the application scenario, and designs the deformation camouflage pattern that meets the activities of the mobile equipment in the area according to the multiple background characteristics such as the forest, grassland, farmland, village, dirt road, cement road and asphalt road passed by the equipment during the travel.

[0086] This embodiment selects a grassland background for generating the dappled pattern of the camouflage, as shown in FIG. 1. Figure 2 As shown in FIG. 1, it is a lawn in a park in Laoshan area of Nanjing, which is acquired by a UAV at a height of 5 meters, and the pixel is 3956*5280. This embodiment selects the grassland acquired at a height of 5 meters as the background for generating the dappled pattern of the camouflage, and the purpose is to reduce the hardware requirements of this method, that is, even if the designer does not have a UAV, a ladder and a camera can be used instead.

[0087] Since the sensitivity of a large equipment, taking a small car on the market as an example, its length is 5 m, its width is 2 m, and the resolution of satellite reconnaissance is 0.1 m, the acquired background picture needs to be segmented into 1000 superpixels, that is, K=S / d 2 =(2*5) / (0.1*0.1)=1000. At the same time, the length-width ratio of the background used for generating the camouflage pattern should be consistent with the target, if not, the background picture needs to be cropped accordingly according to the proportion of the target, and the cropping is to ensure that the generated camouflage texture is consistent with the original background, and there is no compression and stretching phenomenon. In order to ensure that the generated camouflage dapples can be consistent with the target, it needs to be cropped to the size of 2000*5000, which is the same as the length-width ratio of the small car.

[0088] The cropped image is processed by using the superpixel segmentation and fuzzy C-means clustering algorithm to generate the corresponding camouflage pattern, as shown in FIG. 2. Figure 3 .

[0089] Figure 3 is the algorithm for generating the large dappled deformation camouflage pattern by using the superpixel segmentation and fuzzy C-means clustering in the embodiment of the application, wherein Figure 3 (a) in (a) is the original picture, Figure 3 (b) in (b) is the dappled boundary picture generated by superpixel segmentation, Figure 3 (c) in (c) is the three-color clustering picture generated by fuzzy C-means, Figure 3 (d) in (d) is the deformation camouflage dappled pattern obtained by edge extraction on the clustered picture.

[0090] According to the need, the main color of the camouflage is replacedFigure 3 In (c) in the figure, the three colors are replaced by the earth color (RGB value: 230, 192, 161), the green color (124, 128, 64) and the dark green color (68, 77, 54) from dark to light, and a corresponding large spot deformed camouflage pattern is obtained, as shown in Figure 4 .

[0091] As can be seen from Figure 4 , each spot is an irregular spot alone, which meets the irregularity of the deformed camouflage; the minimum size of the camouflage spot designed by the method can also meet the visibility of the spot, and spatial color mixing does not occur at the corresponding reconnaissance distance; any edge of the camouflage basically contains three different colors, which meets the brokenness of the edge, and in theory, the deformed camouflage designed by the method should have good deformed camouflage effect.

[0092] The present application aims to design a small spot camouflage, which needs to re-segment Figure 4 , first, the single-color large spot of the deformed camouflage is extracted by using a mask, and then the small spots in each large spot are generated by using an algorithm combining superpixel segmentation and fuzzy C-means clustering.

[0093] After the design of the large spot of the deformed camouflage is completed, each color spot is assigned a unique label, and the labels of all pixel points in the same color spot are the same, so the label can be used to extract the shape of the large spot of the same color.

[0094] Figure 5 is a background pattern of each color large spot shape extracted by the present application embodiment, Figure 5 (a) in the figure is a grassland background pattern for generating a deformed camouflage spot, Figure 5 (b) in the figure is a grassland background pattern extracted by a mask corresponding to the main color 1, Figure 5 (c) in the figure is a grassland background pattern extracted by a mask corresponding to the main color 2, Figure 5 (d) in the figure is a grassland background pattern extracted by a mask corresponding to the main color 3.

[0095] The size of the small spot needs to be smaller than that of the large spot, so the size of the camouflage unit (superpixel) constituting the small spot also needs to be smaller than that of the large spot, and in the present application, the area of the camouflage unit of the small spot is 1 / 4 of that of the large spot, and therefore, when the deformed camouflage of the large spot is generated, the K value of the superpixel clustering is 1000, and when the small spot camouflage is generated, the K value of the superpixel clustering is 4000.

[0096] Figure 6 (a) in the figure is a grassland background of the main color 1 shape extracted by a mask, Figure 6 (b) in the figure isFigure 6 The superpixel segmentation map in (a) is shown in the image. Figure 6 (c) in the middle is Figure 6 (b) The dichromatic image generated by fuzzy clustering does not include black in the image. Figure 6 (d) in the middle is Figure 6 The dot pattern in (c) is shown.

[0097] Figure 7 (a) in the image represents the grass background in the shape of the main color 2 extracted from the mask. Figure 7 (b) in the middle is Figure 7 The superpixel segmentation map in (a) is shown in the image. Figure 7 (c) in the middle is Figure 7 (b) The dichromatic image generated by fuzzy clustering does not include black in the image. Figure 7 (d) in the middle is Figure 7 The dot pattern in (c) is shown.

[0098] Figure 8 (a) in the image represents the grass background in shape 3, extracted from the mask, representing the primary color. Figure 8 (b) in the middle is Figure 8 The superpixel segmentation map in (a) is shown in the image. Figure 8 (c) in the middle is Figure 8 (b) The dichromatic image generated by fuzzy clustering does not include black in the image. Figure 8 (d) in the middle is Figure 8 The dot pattern in (c) is shown.

[0099] After the small dot pattern of the deformed camouflage is generated, simply replace the colors in the dichromatic clustering diagram with the camouflage main color, and then superimpose the different dots to generate the small dot deformed camouflage.

[0100] When determining the primary colors for the morphing camouflage, only three colors were selected. Each large spot, after generating corresponding smaller spots, required two colors for filling. The filled smaller spots needed to produce good spatial color mixing when observed from a distance. After conducting numerous spatial color mixing experiments, it was found that spatial color mixing is related to the colors themselves; the smaller the color difference between the two colors, the stronger the color mixing ability. In this embodiment, the original color of the large spot is retained as one of the larger areas of the smaller spots. Then, as needed, a color with brightness and small color difference that is closest to the original color of the large spot is selected as the other color of the smaller spots.

[0101] Figure 9 (a), (c), and (e) in the diagram represent large patches of camouflage in a single color. Figure 9 (b), (d), and (f) in the image represent the corresponding small dot camouflage patterns. As can be seen from the image, only... Figure 9 In (b), the larger spots of main color 1 are more noticeable, while the smaller spots are more prominent.Figure 9 The small spots in the two large spots of (d) and (f) in the figure have shown good spatial color mixing effect.

[0102] Figure 10 The effect picture of the three large spot camouflage converted into small spot camouflage is a whole small spot camouflage pattern. The two colors in the earth color spots are relatively obvious, and the different colors in the green spots are difficult to distinguish, the spatial color mixing effect is greatly related to the colors participating in the color mixing. In the case of long distance, the earth color spots will also be spatially mixed, since the two colors in the earth color spots have the same brightness, even if the spatial color mixing effect is not good, the earth color spots also meet the requirements of the main colors of the metamorphic camouflage between other spots, and will not affect the metamorphic effect of the camouflage.

[0103] The colors used in the application are the main colors of a typical background in an eastern region, and the user can replace them with the colors needed by the user.

[0104] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for generating a deformed dappled camouflage based on superpixel segmentation and fuzzy C-means clustering, characterized in that, The method comprises the following steps: Step 1: obtaining a background picture of a natural background; Step 2: cutting the background picture into the same length-width ratio as the target to be camouflaged; Step 3: calculating the number K of superpixel segmentation according to the area of the target and the reconnaissance resolution of the reconnaissance equipment, and performing superpixel segmentation on the background picture; Step 4: replacing other colors with average colors in each superpixel to obtain an average color superpixel pattern; Step 5: clustering the colors of the superpixels into M colors, replacing the colors of the clustered superpixels with the colors of the cluster centers to generate a large spot deformation camouflage pattern of M colors, and replacing the M colors in the pattern with a selected main color of the camouflage to obtain a corresponding large spot deformation camouflage pattern; Step 6: performing edge extraction on the clustered picture to obtain a spot pattern of the large spot deformation camouflage; Step 7: assigning a unique label to each color spot, and the labels of all pixel points in the same color spot are the same, and the same color large spot shape is extracted using the label to obtain a background picture of the large spot shape of each color; Step 8: using an algorithm combining superpixel segmentation and fuzzy C-means clustering to generate small spots in each large spot; Step 9: superimposing the small spot patterns obtained from each large spot to obtain a small spot deformation camouflage pattern.

2. The method according to claim 1, wherein, The background picture of the natural background obtained in step 1 is obtained as follows: A background picture of a natural background is obtained through an image acquisition device, and the picture only contains natural scenery and does not contain artificial targets.

3. The method of claim 2, wherein the method is characterized by, The background picture is cut into the same length-width ratio as the target to be camouflaged in step 2 as follows: The size of the target to be camouflaged is measured, the length-width ratio of the target is calculated, the background picture is cut into the corresponding target ratio, and the cut background picture contains N pixels.

4. The method of claim 3, wherein the method is characterized by, The number K of superpixel segmentation is calculated according to the area of the target and the reconnaissance resolution of the reconnaissance equipment in step 3 as follows: The SLIC algorithm is adopted to segment the cropped background picture into K superpixels with consistent size. The size S of the superpixel is sqrt(N / K), and the size of the spot of the metamorphic camouflage is D is the observation distance; To ensure that there is no spatial color mixing between the spots of the metamorphic camouflage, the minimum value of the spot size d of the metamorphic camouflage is equal to the value of the reconnaissance resolution, the area A of the target to be camouflaged is determined, and the number K of superpixel partitions is calculated = A / d 2 ; The spot size d of the deformation camouflage is organically combined with the number K of superpixel segmentation, and K is adjusted to make the designed spot pattern meet the size requirements of the deformation camouflage spots.

5. The method of claim 1, wherein the method is characterized by, The value of the number M of color clusters in step 5 is M=3.

6. The method of claim 1, wherein the method is characterized by, In step 7, a unique label is assigned to each color spot, and the labels of all pixel points in the same color spot are the same, and the same color large spot shape is extracted using the label to obtain a background picture of the large spot shape of each color as follows: Step 7.1: for each color spot, create a matrix with the same size as the original image as a mask; Step 7.2: find the position of the color spot to be kept in the mask matrix and set its value to 1, and set the values of the other positions to 0; Step 7.3: multiply the background image by the mask, and when the value in the mask matrix is 1, the background image does not change; when the value in the mask matrix is 0, the background image becomes black, thereby extracting the corresponding shape background image.

7. The method of claim 1, wherein the method is characterized by, In step 8, the algorithm combining superpixel segmentation and fuzzy C-means clustering is used to generate small spots in each large spot as follows: Step 8.1, according to the ratio of the small spot camouflage unit area to the large spot camouflage unit, the number K' of superpixel segmentation of the small spot is calculated, and the background picture is superpixel segmented; Step 8.2, in each superpixel of the small spot, the average color is used to replace other colors to obtain the superpixel pattern of the average color of the small spot; Step 8.3, the color of the superpixel of the small spot is clustered into M' colors, the color of the cluster center is used to replace the color of the superpixel of the cluster, M' color small spot deformation camouflage pattern is generated, and the selected main color of the camouflage is used to replace the M' colors in the pattern to obtain the corresponding small spot deformation camouflage.

8. The method of claim 7, wherein the method is characterized by, The value of the number K' of superpixel segmentation of the small spot in step 8.1 is 4 times the number K of superpixel segmentation of the large spot, that is, K' = 4K.

9. The method of claim 7, wherein the method is characterized by, The value of the number M' of color types of the superpixel of the small spot in step 8.3 is M' = 2.

10. The method of claim 1, wherein the method is based on superpixel segmentation and fuzzy C-means clustering. Step 9, the small spot patterns obtained by each large spot are superimposed to obtain a small spot deformation camouflage pattern, which is as follows: After the small spot pattern of the deformation camouflage is generated, the main color of the camouflage is used to replace the color in the two-color clustering diagram, and different spots are superimposed to generate a small spot deformation camouflage pattern.

Images