基于背景拼接与纹理模板的全自动迷彩图案设计

Fully automated camouflage pattern design based on background stitching and texture templates

为提高迷彩伪装图案的设计效率和环境适应性,利用背景拼接与纹理模板融合进行迷彩伪装图案的全自动化设计。通过将输入的若干背景图像自动组合为单张拼接图像,采用均值聚类法提取出背景拼接图像的主色;提出4种迷彩纹理模板自动设计方法(多圆形随机分布法、WGN傅里叶频谱法、肌理图像生成法及分层云彩法),再对所得迷彩纹理模板进行色彩聚类与色彩替换,得到最终的迷彩伪装图案。为验证设计的有效性,针对丛林虚拟场景自动生成了4种伪装图案,并进行了伪装效果的主观和客观评价。结果表明,上述方法仅需17.0~71.0 s即可完成针对目标背景的迷彩图案设计,且最优方案为基于多圆形随机分布法纹理模板的迷彩图案;本文方法相对于普通迷彩伪装,其主观评价搜索时间增加了8.1%,基于Positioning and Focus Network(PF-Net)模型的客观评价发现概率降低了45%。

Objective In the military,the ability to reponse quickly to the battlefield environment can determine the success or failure of a battle.Therefore,It is need ed to improve the design speed of camouflage patterns and the ability to adapt to the target environment to enhance the camouflage effect of patterns and thus improve the military level of the army.Method First,we automatically combined the background environment dataset into a stitching map and extracted the primary color of the background stitching map by mean clustering method;then,we proposed four automatic design methods for camouflage texture templates,including multi-circular random distribution method,WGN Fourier spectrum method,texture image generation method,and layered cloud method.Finally,after the mean clustering of camouflage texture templates,the camouflage pattern was obtained by replacing the template colors with the primary colors.Results Experiments were conducted to construct virtual scenes and a dataset of target background images was created.We stitched the images in the dataset into a single large stitched image.The stitched image was detected based on a color histogram in RGB color space,and a cluster K value of 6 was determined.Six colors were extracted from the eight virtual jungle terrain scenes as primary colors,and four different camouflage patterns were generated based on a texture template method.The experimental time was short,and the design rate was high.In order to reduce the impact of environmental and human factors,camouflage assessment has been relatively carried out using templates based on multi-circular random distribution method,templates based on WGN Fourier spectrum method,templates based on texture image generation method,and templates based on layered cloud method as camouflage targets,and non-environment specific jungle camouflage(FLECKTARN-style jungle camouflage)as a reference.In the subjective evaluation experiments,the experimental results.Each test set had a 100%probability of discovery and different average search times,with the test set of camouflage patterns designed based on the multi-circular random distribution method having the longest average search time of 1.5614 s.The average search times of all four camouflage patterns designed using this paper′s method were greater than those of the reference pattern.Among them,the camouflage search time increased by 8.1%for the template based on the multi-circular random distribution method,1.1%for the template based on the WGN Fourier spectroscopy method,2.4%for the template based on the texture image generation method,and 3.7%for the template based on the layered cloud coloring method.In the objective evaluation experiments,the experimental results were shown.The average search time of the PF-Net network for the test set images was the same for both,0.04 s,but the probability of discovery was different.The detection probability of the test sets designed according to the fully automated camouflage pattern design method based on background stitching and texture templates were both lower than that of the reference target,with the detection probability of the template camouflage based on the multi-circular random distribution method being 45%lower than the detection probability of the reference target.Conclusion For automatic camouflage to achieve excellent results,it is necessary to react quickly to generate camouflage colors and textures according to changes in the target environment.This paper proposes a fully automatic camouflage pattern design method based on background splicing and texture templates with a high design rate,rapid response to the target environment,and good camouflage effect according to the modern army′s demand for camouflage combined with computer technology.Through the subjective and objective evaluation of camouflage detection,the camouflage effect of the camouflage pattern is evaluated with the search time and detection probability as the primary indexes.The feasibility and effectiveness of the method are verified,and the generated pattern has a good camouflage effect.The camouflage design method in this paper provides various ideas for constructing camouflage design and achieves a more accurate and convenient camouflage pattern design for target area environments.