CAJ | 학술논문

合成孔径雷达(SAR)图像中固有的相干斑噪声，严重影响了图像分割算法性能。为了改善SAR图像分割质量，本文提出了一种联合PM扩散模型和各向异性MRF模型的图像分割方法。首先对传统PM扩散模型的扩散系数进行简化和近似，限制模型的解的唯一；然后使用改进后的模型对原始SAR图像进行非线性扩散，在抑制噪声的同时保持图像结构细节；继而，通过在标记场势能函数中引入观测数据灰度信息，将经典的基团势能改进为基于灰度加权的各向异性势能，提高边缘像素和图像奇异点的分割准确率。实验表明，本文算法的分割结果区域连通性更好，边缘轮廓分割更精细。
합성공경뢰체(SAR)도상중고유적상간반조성，엄중영향료도상분할산법성능。위료개선SAR도상분할질량，본문제출료일충연합PM확산모형화각향이성MRF모형적도상분할방법。수선대전통PM확산모형적확산계수진행간화화근사，한제모형적해적유일；연후사용개진후적모형대원시SAR도상진행비선성확산，재억제조성적동시보지도상결구세절；계이，통과재표기장세능함수중인입관측수거회도신식，장경전적기단세능개진위기우회도가권적각향이성세능，제고변연상소화도상기이점적분할준학솔。실험표명，본문산법적분할결과구역련통성경호，변연륜곽분할경정세。
Speckle noise that is inherent in SAR imagery makes the traditional image segmentation methods inefficient. To improve the performance, a novel method combining PM diffusion model and the anisotropic MRF model is proposed in this paper. First, to restrict the uniqueness of the solution, the traditional PM diffusion model is simplified using mathematic approximation. Then the modified diffusion model is employed to smooth the speckle noise of SAR image. Second, the filtered image is segmented using MRF model-based method, in which the original intensity information is introduced into the computation of potential function, i. e., the clique potential is weighted by local difference of intensity values, thus the anisotropic potential is achieved. Experimental results on real SAR imagery demonstrate that the proposed method segment the image more uniformly and the contour position is more exact than the traditional method.