论文部分内容阅读
高斯混合模型(GMM)聚类算法近年来广泛应用于图像分割领域。但在SAR图像分割中,由于忽略了图像像素间的空间相关性,使其对相干斑噪声十分敏感。提出一种基于区域的GMM聚类算法,它将空间相关性引入聚类分类中,利用分水岭分割得到基本同质区域,计算区域的灰度均值作为GMM聚类算法的输入样本,将聚类特征从像素水平提升到区域水平,减少了噪声对分割结果的影响;并将自身反馈机制引入期望最大化(EM)算法中,进一步提高了GMM模型参数估计的精度。还对合成图像和真实SAR图像进行了分割实验,结果表明新算法可有效地提高分割的准确性。
Gaussian mixture model (GMM) clustering algorithm has been widely used in image segmentation in recent years. However, SAR image segmentation, due to neglect the spatial correlation between image pixels, make it very sensitive to speckle noise. This paper proposes a region-based GMM clustering algorithm, which introduces spatial correlation into the clustering classification, divides the watershed into basic homogeneous regions and calculates the gray mean value of the region as the input sample of the GMM clustering algorithm. The clustering feature The performance of GMM model is further improved by introducing self-feedback mechanism into expectation maximization (EM) algorithm, which improves the segmentation result from pixel level to region level. The segmentation experiment of synthetic image and real SAR image is also carried out. The results show that the new algorithm can effectively improve the accuracy of segmentation.