基于NSCT-Gabor特征和脉冲耦合神经网络的SAR图像分割  

A Segmentation Algorithm for SAR Images Based on NSCT-Gabor Characteristics and PCNN

作　　者：吴俊政[1] 严卫东[1] 倪维平[1] 边辉[1] 张晗[1] 

机构地区：[1]西北核技术研究所,西安710024

出　　处：《电光与控制》2015年第4期95-99,共5页Electronics Optics & Control

摘　　要：针对SAR图像目标的精确分割问题,利用非下采样轮廓波变换(NSCT)和Gabor滤波器分别提取图像特征,然后采用脉冲耦合神经网络(PCNN)对目标区域进行增强,提出了一种分割算法。分别对图像进行NSCT分解和Gabor滤波,对NSCT域的高、低频子带系数构造一个特征图,对Gabor滤波的不同尺度构造对应的特征图,对所获取的各个特征图用PCNN进行目标增强,最后对增强的特征图进行合理合并与分割。利用MSTAR SAR数据库中各种干扰强度下的图像进行了实验,结果表明,相比于模糊C均值、马尔可夫随机场等常见的分割算法,所提出的算法分割结果更为准确,同时受噪声干扰更小。A segmentation algorithm was proposed by using Nonsubsampled Contourlet Transform（ NSCT）and Gabor filter to extract characteristics of images respectively and using Pulse Coupled Neural Networks（ PCNN） to enhance the target areas. Characteristic figures were constructed for the high and low frequencies of NSCT and corresponding characteristic figures were also constructed for the Gabor filters. All the characteristic figures were enhanced by PCNN. Then,the enhanced figures were integrated and segmented reasonably. Images in MSTAR SAR data library under different jamming intensities were selected for experiment. The results indicated that： Compared with the common algorithms such as FCM and the algorithm based on Markov random field,the proposed algorithm can realize more accurate segmentation for SAR images and has strong immunity from interferences.

关 键 词：SAR图像 图像分割 非下采样CONTOURLET变换 GABOR滤波器 特征图 MSTAR图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]