基于马尔可夫随机场的快速图象分割

Markov Random Field Based Fast Segmentation

根据卫星遥感图象的特点 ,讨论了基于马可夫随机场的图象分割方法 ,建立了相应的基于马可夫随机场的图象分割模型 ,以实现复杂遥感图象的快速分割 ,并由此将图象分割问题转化成图象标记问题 ,进而转化成求解图象的最大后验概率估计的问题 .虽然传统的模拟退火算法 (SA)能达到后验概率的全局最大 ,但是时间复杂度太高 ,实际分割中经常采用次优算法 .文中还引进了一种基于博弈理论的决定性退火算法 (GSA)和一种基于竞争理论的算法 (CA) ,取得了快速分割图象的效果 .试验证明 ,该两种算法完全可应用于复杂遥感图象的快速分割 .

In this paper,the segmentation based on Markov Random Field (MRF) is discussed to fulfill the fast segmentation of complex remote sensing image. Using this method, the cotton estimation model and the extraction of cotton areas from satellite image are realized and remote sensing cotton estimation system is constructed. According to the characteristics of the remote sensing image,the image segmentation model based on MRF is established.The problem of image segmentation can be converted to the problem of symbolizing,and finally converted to the solution of Maximum A Posterior (MAP), if the method of MRF is used. For obtaining the solution of MAP, the algorithm of simulated annealing (SA) can find the global optimum,but it requires a large amount of computation. So sub optimal algorithms are often used. In the article, the decisive algorithm based on game theory and the algorithm based on competition theory are both introduced. Moreover the competition algorithm(CA) is improved largely. These two algroithms reduce the complexity from different way. The experiments indicated that they could be used in the segmentation of complex remote sensing image effectively. In the system constructed by the method, the cotton areas are extracted with high precision from satellite image.