为提高遥感图像水域分割的准确度, 结合高分率遥感图像中水域与背景纹理复杂度差异较大的特点, 将图像熵引入到CV模型中, 提出两种图像熵主动轮廓模型用于高分辨率遥感图像的水域分割。其中, 针对水域纹理相对简单的遥感图像, 在CV模型中引入零水平集内的图像熵而构成局部图像熵主动轮廓模型, 可以有效降低背景中灰度值与水域近似的区域发生误分, 从而提高水域分割的准确度; 针对水域纹理相对复杂的遥感图像, 在CV模型中同时引入零水平集内外图像熵而构成全局图像熵主动轮廓模型, 改进了水平集函数进化过程中对灰度信息的依赖, 并能使零水平集进化到全局最优, 进一步提高了遥感图像中水域分割的准确度。针对高分辨率遥感图像中的湖泊、河流和海域分割对比实验结果表明：局部图像熵主动轮廓模型的分割精确率分别为90.1%、81.5%和93.6%, F值分别为0.94、0.885和0.96; 全局图像熵主动轮廓模型的分割精确率分别为94.5%、85.3%、94.9%, F值分别为0.956、0.895、0.967。本文提出的两种图像熵主动轮廓模型均能有效减小背景误分, 提高了遥感图像水域分割的准确度。

In order to improve the accuracy of water area segmentation in high resolution remote sensing image, the image entropy was introduced into CV model because there was a quite difference of texture complexities between water area and background, and two active counter models based on image entropy were proposed in this paper. The image entropy of inside zero level set was adopted in CV model and forms a local image entropy active counter model (LIEACM). This model effectively reduced the incorrect segmentation of background where the gray value approximated to the water area with low texture complexity. For remote sensing image of water area with high texture complexity, the global image entropy active counter model (GIEACM) was proposed, in which, the image entropy of inside and outside of zero level set were employed in CV model simultaneously. GLEACM modifies the fact that the level set function evolution depends on gray value, and the zero level set cald evaluate to the global optimal value. The experiments on segmentation the lake, river and sea validate that the segmentation precisions of LIFACM are 90.1%, 81.5% and 93.6%, respectively, the F-measures are 0.94, 0.885 and 0.96, respectively; and for GLEACM, the segmentation precisions are 94.5%, 853% and 94.9%, respectively, the F-measures are 0.956, 0.895 and 0.967, respectively. The two image entropy active contour models proposed by this paper improve the water area segmentation accuracy in remote sensing image effectively.