四叉树与多种活动轮廓模型相结合的遥感影像水边线提取方法

A Waterline Extraction Method from Remote Sensing Image Based on Quad-tree and Multiple Active Contour Model

通过对测地线活动轮廓(GAC)模型、Chan-Vese(CV)模型、局部二值拟合(LBF)模型的分析,将基于边缘和区域的活动轮廓模型以及基于四叉树的影像分割方法有机结合,提出了一种基于四叉树和多种活动轮廓模型的水边线提取方法。该方法首先对影像进行四叉树分割,为模型演化提供初始轮廓;然后利用CV模型的全局区域图像统计信息和LBF模型的局部区域图像统计信息构造新的符号压力函数,利用改进的符号压力函数代替GAC模型的边界停止函数,有效地改善了GAC模型提前停止演化和过度演化的问题;最后采用二值选择和高斯滤波正则化水平集方法(SBGFRLS)进行演化,避免了重新初始化和规则化,提高了水平集演化的效率。试验结果表明该方法对于包括弱边缘和严重凹陷边缘的水边线提取效果均良好,具有亚像素提取精度,并且提取速度快、稳定性好。

After the characteristics of geodesic active contour model（ GAC）,Chan-Vese model（ CV） and local binary fitting model（ LBF） are analyzed,and the active contour model based on regions and edges is combined with image segmentation method based on quad-tree,a waterline extraction method based on quad-tree and multiple active contour model is proposed in this paper. Firstly,the method provides an initial contour according to quadtree segmentation. Secondly,a new signed pressure force（ SPF） function based on global image statistics information of CV model and local image statistics information of LBF model has been defined,and then,the edge stopping function（ ESF） is replaced by the proposed SPF function,which solves the problem such as evolution stopped in advance and excessive evolution. Finally,the selective binary and Gaussian filtering level set method is used to avoid reinitializing and regularization to improve the evolution efficiency. The experimental results show that this method can effectively extract the weak edges and serious concave edges,and owns some properties such as sub-pixel accuracy,high efficiency and reliability for waterline extraction.