基于非线性形状先验的高分辨率遥感图像目标提取方法

Nonlinear shape prior-based object extraction in high resolution remote sensing images

对已知目标形状,利用核主成分分析方法非线性建模,并根据高分辨率遥感图像特点,提出一种新的融入形状先验、图像边缘、颜色以及阴影信息的目标提取方法.该方法构造了基于活动轮廓模型的能量函数,并通过迭代的全局最优化方法最小化,实现对目标的准确分割提取.实验结果表明,该方法不仅能准确高效分割提取目标区域,而且能抵制背景噪声干扰,具有很强的鲁棒性和实用价值.

In this study, given shape templates of some object, we model them by using kernel principal component analysis and then propose a new object extraction method for high resolution remote sensing images, which integrates shape prior and several image appearance information, including edge, color, and shadow. Based on active contour model, a new energy function is constructed and minimized through an iterative global optimization method to get the accurate segmentation results. Experimental results show that our method has high efficiency, high accuracy, and the robustness with respect to various background disturbances.