欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 图像与视频处理 • 上一篇    下一篇

自动确定类别数的 RJMCMC+SA 图像 分割算法研究

  

  1. (1. 宁夏大学物理与电子电气工程学院，宁夏 银川 750021；
     2. 宁夏沙漠信息智能感知自治区重点实验室，宁夏 银川 750021)
  • 出版日期:2019-12-31 发布日期:2020-01-20
  • 基金资助:
    国家自然科学基金项目(41561087)

The Research about RJMCMC+SA Image Segmentation Algorithm to Automatically Determine the Number of Categories

  1. (1. School of Physics and Electronic-Electrical Engineering, Ningxia University, Yinchuan Ningxia 750021, China; 
    2. Ningxia Desert Information Intelligent Perception Autonomous Region Key Laboratory, Yinchuan Ningxia 750021, China)
  • Online:2019-12-31 Published:2020-01-20

PDF (PC)

75

摘要： 在保证遥感图像分割模型复杂性、分割精度的情况下，自动确定分割类别数是一 个重点问题。为此，结合可逆跳马尔科夫蒙特卡洛和模拟退火理论(RJMCMC+SA)构建了图像 分割算法。通过高斯曲率滤波(GC)对图像进行几何平滑处理，依据贝叶斯理论形式化非线性回 归模型中的参数变量从而建立后验概率分布，利用 RJMCMC 算法实现该后验概率分布并构建 概率转移核，结合 SA 算法加速概率转移核收敛，确定分割算法中径向基函数的个数和参数， 完成类别数自动确定和图像全局性分割。在全色遥感图像和伯克利大学实验数据库图像上，分 别与 4 种径向基函数分割模型实验对比，数据分析表明，该算法不仅在复杂性和精确度上取得 很好的平衡，而且能够自动确定图像类别数。

关键词: 高斯曲率滤波, 后验概率, RJMCMC+SA, 模型选择方法, 混淆矩阵

Abstract:  Based upon the premise of ensuring the complexity and accuracy of remote sensing image segmentation model, it is crucial to automatically determine the number of segmentation categories. An image segmentation algorithm is constructed in this paper by combining reversible jump Markov Chain Monte Carlo and simulated annealing (RJMCMC+SA). The image is smoothed geometrically by Gauss curvature filtering (GC), and the posterior probability distribution is established by formalizing the parameters in the nonlinear regression model based on Bayesian theory. And then RJMCMC algorithm is used to accomplish the posterior probability distribution and construct the probability transfer core. After that SA algorithm is used to accelerate the convergence of the probability transfer kernel to determine the number and parameters of the radial basis function in the segmentation algorithm and complete the automatic determination of the number of categories and image global segmentation. Finally, the segmentation algorithm is compared with four RBF segmentation models in panchromatic remote sensing images and Berkeley University experimental database images. The data analysis shows that the algorithm not only strikes a good balance in complexity and accuracy, but also automatically determines the number of image categories.

Key words: Gauss curvature filtering, posterior probability, RJMCMC+SA, model selection method, confusion matrix