基于Chan-Vese模型的SAR图像分割

摘要

Due to strong speckle noise in synthetic aperture radar (SAR) image, the Chan-Vese model level set segmentation method produces a lot of false segmentation. Meanwhile, the level set has disadvantages of large amount of computation and slow segmentation velocity. There fore , a new internal force term- distance regularized term is introduced to create an improved curve evolution model based on the Chan-Vese model. The model avoides the periodic updates of level set function and has a longer time step. So the segmentation speed is speeded up, and the anti-noise capability is enhanced. Then, the model is tested by processing the synthetic im age and real SAR images. By comparison, the improved model has higher numerical accuracy and faster division speed. As for the image with strong noise, using the enhanced Lee filter can further improve the speed and effect of the segmentation model. The result shows that the im proved Chan-Vese model can complete SAR image segmentation rapidly and efficiently with high robustness.%由于SAR图像存在较强的斑点噪声,使用Chan-Vese模型水平集分割方法会产生很多误分割.同时,水平集解法存在计算量大、分割速度慢的问题.在Chan-Vese模型基础上,增加新的内能项——距离正则项,得到了一种改进的曲线演化模型.避免了水平集函数的周期性更新,具有更大的迭代步长,从而加快分割速度,并且提高Chan-Vese模型的抗噪性.对该模型采用人工合成图像和真实SAR图像进行分割实验,通过比较,可看出改进模型具有较高的数值精度和较快的分割速度.对于噪声很强的图像,使用增强Lee滤波进行预处理,可以进一步提高改进模型的分割速度和效果.实验结果表明:改进Chan-Vese模型能高效快速地完成SAR图像分割,具有较高的抗噪性.