一种基于k均值的多相位水平集遥感图像分割方法

针对遥感图像的特点,本文提出了一种基于K-均值与改进的多相位水平集模型结合的新方法。相比于传统的水平集模型,改进模型在能量函数中考虑了图像的面积、梯度信息和边缘检测。图像的梯度信息可以克服分割中存在的边缘定位的不准确,边缘检测可以在曲线衍化过程中更好的保持边缘信息。为了加快边缘的收敛速度,避免陷入局部最优,本文提出先对图像进行中值滤波来平滑图像和消除部分噪声,然后利用K均值进行聚类得到明显的特征差异。接着用Sobel算子进行梯度重建,然后用改进的多相位水平集模型进行分割。实验结果显示本文的算法对于遥感图像的分割在时间和精度上都有较好的效果。

A novel level set method for remote sensing image based on K-means algorithm

Due to the characteristic of remote sensing image, a novel method based on K-means algorithm and improved multi-phrase level set model is proposed. Compare to Classical multi-phase C-V model, improved model considered the region area, gradient information and edge detection. The use of gradient information can overcome the inaccurate edge localization defects in im-age segmentation. The edge detection was used to keep the boundary information better in the evolution process. In order to accelerate the contour’s convergence speed and avoid it trapping into local optimal. Firstly, a median filtering was applied to smooth the origi-nal image, so it reduced part of noise. Secondly, use K-means to gain more obvious differences of characteristics. Then, the reconstruc-tion of gradient is obtained by using Sobel operator. Finally, segmentation result is achieved by using an improved method of multi-phase level set image segmentation. Experimental results show that the proposed approach has advantages in speed in compari-son with the classical multi-phase C-V model on the application of remote sensing image segmentation.