SAR图像河流提取的主动轮廓模型的稳健估计算法

针对现有主动轮廓模型无法精确提取SAR图像中河流的难题,提出了一种结合L1范数和拉普拉斯能量的主动轮廓模型。首先,将Chan-Vese(CV)模型中L2范数形式的外部能量约束项替换为L1范数形式的外部能量约束项,得到新的能量泛函;其次,提出了一种基于拉普拉斯核函数的外部能量约束项,并将其添加到上述能量泛函中,同时赋予两种外部能量约束项不同的调节系数;最后,引入曲线内外区域像素灰度绝对中位差的均值替代模型中的常数曲线内外能量权值,以得到完整的提出模型。针对实际SAR图像进行河流提取,结果表明:与现有主动轮廓模型相比,本文提出的模型在河流提取准确性和提取效率两方面具有明显优势。

Robust estimation algorithm of active contour model for river extraction in SAR images

To deal with the problem that the existing active contour models are unable to extract rivers in SAR images accurately, this paper presents an new active contour model with L₁ norm and Laplacian energies. First, the external energy constraint in form of the L₂ norm in the CV model is replaced by the external energy constraint in form of the L₁ norm and then the novel energy functional is obtained. Second, an external energy constraint based on the Laplacian kernel function is proposed and added to the above energy functional. Meanwhile, the different adjustment coefficients are assigned to these two external energy constraints. Finally, the mean of median absolute deviations of pixel grayscale values inside and outside the curve is introduced to replace the constant energy weights inside and outside the curve of the model and then the completed proposed model is developed. River extraction is carried out on real SAR images and the results reveal the superiority of the proposed model on both accuracy and efficiency of the river extraction, when compared with the existing active contour models.