【摘要】 随着超像素算法的发展,简单线性迭代聚类算法(SLIC)简单和良好的分割效果被广泛应用,主要应用在林地信息提取、目标跟踪、生物医学图像等领域。但针对简单线性迭代聚类算法在纹理较为复杂的遥感影像分割中的不足,提出了一种改进SLIC的超像素分割算法(M-SLIC)。首先使用均值漂移算法对预处理后的LandSat8遥感影像进行均值漂移,再结合均匀模式+旋转不变的线性反投影算法(LBP)和SLIC算法,对遥感影像进行超像素分割处理。结果表明:M-SLIC超像素分割算法分割遥感影像时相比较SLIC算法在边缘召回率上提高了1.5%,在欠分割错误率上降低了2%,M-SLIC算法能够更好地贴合遥感影像边缘,对于颜色相近的地物分割的欠分割错误率更低,更适合纹理较为复杂的林地遥感影像。更多还原

【Abstract】 Along with the development of the super pixel algorithm, simple linear iterative clustering(SLIC) algorithm with low time complexity and good segmentation effect is widely used, mainly in forest land information extraction, target tracking, in areas such as biomedical image plays an important role. For SLIC in the complex remote sensing image texture segmentation, we proposed an improved SLIC super pixel segmentation algorithm(M-SLIC algorithm). First, the Mean Shift algorithm was used to carry out Mean Shift of the pre-processed LandSat8 remote sensing image, and then the super-pixel segmentation of remote sensing image was carried out by combining the uniform Pattern+rotation invariant LBP(linear back projection) algorithm and SLIC algorithm. In order to verify the segmentation effect, remote sensing images with complex texture were selected for experiment. By contrast experiment analysis, comparedwith the proposed M-SLIC super pixel SLIC, the segmentation algorithm for remote sensing image segmentation algorithm on the edge of the recall rate was increased by 1.5%, and the under segmentation error rate was reduced by 2%. the M-SLIC algorithm can better fit the remote sensing image edge, for similar color object segmentation under segmentation error rates are lower, more suitable for more complex forest land remote sensing image texture.更多还原