基于复contourlet域隐马尔科夫树模型的海面溢油合成孔径雷达图像相干斑抑制

Marine spill oil SAR images despeckling based on hidden Markov tree model in complex contourlet domain

海面溢油SAR图像中的相干斑噪声严重影响了后续的图像分割、特征提取和分类。为了更有效地抑制海面溢油SAR图像相干斑,文中提出了一种基于复contourlet域隐马尔科夫树模型的海面溢油SAR图像相干斑抑制方法。首先对观测图像取对数并进行复contourlet变换;然后在复contour-let域中用隐马尔科夫树模型对相邻尺度间的带通方向子带系数进行建模,并依据贝叶斯最小均方误差准则估计无噪系数;最后进行逆复contourlet变换和指数变换,得到相干斑抑制后的图像。大量实验结果表明,与Lee、Kuan、Frost及Gamma Map等4种经典滤波方法以及小波域和contourlet域隐马尔科夫树模型方法相比,文中方法从主观视觉和客观定量评价两方面来看综合性能更为优越,是一种行之有效的SAR遥感图像海面溢油检测的预处理方法。

The presence of speckle noise in the marine spill oil SAR images seriously affects the follow--up image segmentation, feature extraction and classification. To suppress the speckle in the marine spill oil SAR images more effectively, a method of reducing the speckle noise in the marine spill oil SAR images based on the hidden Markov tree model in complex Contourlet transform domain is proposed in this paper, firstly, the observed image is taken the logarithm and the complex contourlet transform is performed. Then the hidden Markov tree model is adopted to a model the band pass directional subband coefficients between adjacent scales in complex contourlet domain. Moreover, the denoised coefficients are estimated according to Bayes minimum mean square error criterion. Finally, the inverse complex contourlet transform and the exponential transform are performed to obtain the despeckled image. A large number of experimental results show that, compared with four classical filtering methods such as Lee filter, Kuan filter, Frost filter and Gamma Map filter, and the methods based on the hidden Markov tree model in wavelet or contourlet transform domain, the proposed method in this paper has superior comprehensive performance according to subiective visual and objective quantitative evaluation. It is an effective preprocessing method of marine spill oil detection based on SAR remote sensing images.