Current Issue Cover

谌华,郭伟,闫敬文(中国科学院国家空间科学中心;汕头大学 电子工程系)

摘 要
摘 要:目的 针对传统Grab Cut算法需要人工交互操作,无法实现SAR图像的自动分割,且算法只利用到纹理或边界信息中的一种、方式单一的问题,提出了一种综合利用纹理和边界信息的改进Grab Cut算法,实现对SAR图像目标的自动分割。方法 首先将其他格式的彩色或灰度SAR图像转化为24bit的位图,采用图形理论对整幅 SAR图像建模,根据最大流算法找到描述图的能量函数最小的割集从而分割出目标区域;然后采用中值滤波抑制相干噪声,最后再通过邻域生长算法滤除图像斑点和小目标的干扰从而达到目标边界的连接,实现自动对SAR图像中的目标进行分割。结果 在64位Window 7环境下采用Matlab R2014处理平台,对楼房、车库、大树、汽车群等4幅SAR图像进行目标分割实验,特征目标被自动分割出来,图像中的背景杂波、目标阴影和干扰小目标被有效地去除。结论 综合利用纹理和边界信息能够有效抑制相干噪声,去除图像斑点和小目标的干扰,从而达到目标边界的连接,实现对SAR图像目标的自动分割。实验结果表明,本文算法可以满足工程化应用要求,自适应性强、算法分割精度高,且具有较好的鲁棒性。
A Synthetic Aperture Radar Image Target Segmentation Method Based on Boundary and Texture Information

Chen Hua,Guo Wei,Yan Jingwen(Center for Space Science and Applied Research, Chinese Academy of Sciences;Shantou University)

Objective Aiming at problems of traditional Grab Cut algorithm which requires artificial interaction, target of SAR image cannot be segmented automatically; and one of the information, texture or boundary, used by Grab Cut algorithm, which is too single. The advanced Grab Cut algorithm is proposed which makes use of two kinds of information, texture and boundary, to achieve target segmentation automatically. Method Firstly, the algorithm of this paper transforming the color or gray SAR image of other format into the bitmap of 24bit, modeling whole SAR image by using graphical theory and finding a cut set with the smallest energy function describing the graph according to the maximum flow algorithm for segmenting the target region; Then, the coherent noise is suppressed by using median filter; Finally, the neighborhood growth algorithm is used to filter out the interference between the image spot and the small target to achieve the connection of the target boundary, and the target in the SAR image is automatically segmented. Result In the 64 - bit window 7 environment, Matlab R2014 processing platform is adopted, four SAR images such as buildings, garages, trees, and car groups are tested for target segmentation. The characteristic targets are automatically segmented, and the background clutter, target shadows, and small interfering targets in the images have been effectively removed. Conclusion The comprehensive use of texture and boundary information can effectively suppress coherent noise, remove the interference of image spots and small targets so as to achieve the connection of target boundaries and realize the automatic segmentation of SAR image targets. The experimental results show that the algorithm can meet the requirements of engineering application, has strong adaptability, high segmentation accuracy and good robustness.