多水平集演化的高分辨率遥感影像分割

Multiple Level Sets Segmentation for High Resolution Remote Sensing Imagery

提出多水平集演化的非监督高分辨率影像分割模型,避免传统多水平集方法中的区域重叠问题。提出新的水平集函数重初始化技术加快曲线演化,该模型缓解面向像元分割中的"椒盐效应"及面向对象分割中尺度选择不佳造成的过/欠分割现象。

The paper presents an unsupervised segmentation model based on multiple level sets evolution for high resolution satellite imagery. The model can remove ambiguities occurring to multiple levels sets methods. In the numerical implementation, a novel re-initialization technique for level set functions is suggested to accelerate curves evolution steadily. Experiments show that our method can alleviate ＂pepper and salt effect＂ compared with pixel- based method, and to lessen over- and under-segmentation compared to object-based methods.