多尺度曲波分解下的可变类SAR图像分割

发布时间：2018-05-05 01:36

本文选题：可变类SAR图像分割 + 曲波变换　；参考：《信号处理》2017年08期

【摘要】：为了实现SAR图像的可变类分割,本文提出了一种基于区域的多尺度可变类分割方法。首先,利用曲波变换对SAR图像进行多尺度分解,获取多尺度曲波系数;然后按尺度由粗-细次序,利用曲波逆变换对各尺度曲波系数进行重构,获取各尺度分解图像,进而获得多尺度分解图像。在此基础上,利用规则划分技术划分图像域;然后利用Gamma分布及马尔科夫随机场(Markov Random Field,MRF)模型建立基于区域的特征场模型及标号场模型;假设图像类别数为随机变量,并服从Poisson分布;并在贝叶斯理论框架下建立基于区域的多尺度可变类分割模型。最后,利用可逆变马尔可夫链蒙特卡罗(Reversible Jump Markov Chain Monte Carlo,RJMCMC)算法,实现该模型求解;在求解过程中,按尺度由粗-细次序,将当前尺度分割解作为下一低尺度分割的初始解,以细尺度的分割解作为最终分割结果。利用提出方法对模拟及真实SAR图像进行可变类分割实验,通过其实验结果验证提出方法的可行性及有效性。
[Abstract]:In order to realize the variable class segmentation of SAR image, a region based multi scale variable class segmentation method is proposed in this paper. Firstly, the multiscale SAR image is decomposed by Qu Bo transform to obtain the multiscale Qu Bo coefficient, and then the Qu Bo coefficients of each scale are reconstructed by Qu Bo inverse transform according to the coarse-fine order of scale, and the decomposed images of each scale are obtained. Then the multi-scale decomposition image is obtained. On this basis, the image domain is divided by rule partitioning technique, and then the area-based feature field model and label field model are established by using the Gamma distribution and Markov random field Random MRF model, and the number of image categories is assumed to be a random variable. The model of multi-scale variable class segmentation based on region is established under the framework of Bayesian theory. Finally, the model is solved by using the invertible Markov chain Monte Carlo reducible Jump Markov Chain Monte realizable Jump Markov Chain Monte (RJMCMC) algorithm, in which the current scale segmentation solution is regarded as the initial solution of the next low scale segmentation according to the coarse-fine order of the scale. The segmentation solution of fine scale is used as the final segmentation result. The proposed method is used to perform variable class segmentation experiments on simulated and real SAR images. The experimental results show that the proposed method is feasible and effective.
【作者单位】：辽宁工程技术大学测绘与地理科学学院遥感科学与应用研究所;
【基金】：国家自然科学基金项目(41271435);国家自然科学基金青年科学基金项目(41301479) 辽宁工程技术大学研究生教育创新计划(YB201604)
【分类号】：TN957.52

【相似文献】