一種采用高斯隱馬爾可夫隨機(jī)場模型的遙感圖像分類算法
A remotely sensed image classification algorithm based on gaussian hidden markov random field model
-
摘要: 該文研究了無監(jiān)督遙感圖像分類問題。文中構(gòu)造了圖像的隱馬爾可夫隨機(jī)場模型(HiddenMarkov Random Fleid,HMRF),并且提出了基于該模型的圖像分類算法。該文采用有限高斯混合模型(Finite Gaussian Mixture,FGM)描述圖像像素灰度的條件概率分布,使用EM(Expectation-Maximization)算法解決從不完整數(shù)據(jù)中估計(jì)概率模型參數(shù)問題。針對遙感圖像分布的不均勻特性,該文提出的算法沒有采用固定的馬爾可夫隨機(jī)場模型參數(shù),而是在遞歸分類算法中分級地調(diào)整模型參數(shù)以適應(yīng)區(qū)域的變化。實(shí)驗(yàn)結(jié)果表明了該文算法的有效性,分類算法處理精度高于C-Means聚類算法.。
-
關(guān)鍵詞:
- 遙感; 模式分類; 馬爾可夫隨機(jī)場; EM算法
Abstract: The problem of unsupervised classification of remotely sensed image is considered in this paper. A Hidden Markov Random Field (HMRF) model is built and a new image clas-sification algorithm based on the HMRF model is presented to the remote sensing application. In the algorithm, the Finite Gaussian Mixture (FGM) model is used to describe the density function of the image pixel intensity, the Expectation Maximization (EM) algorithm is used for the HMRF model parameters under the incomplete data condition, and MAP (Maximum A Posteriori) method is used to estimate the image class label. As the MRF model with fixed parameters does not fit the real remotely sensed image very well, this paper adjusts the MRF models parameters during the classification procedure. The novel image classification method gives a more accurate and more robust image classification. -
Tulkel Derin, et al., Modeling and segmentation of noisy and textured images using Gibbs random fields, IEEE Trans. on PAMI., 1987, PAMI-9(1), 39-55.[2]T.N. Pappas, An adaptive clustering algorithm for image segmentation, IEEE Trans. on Signal Processing, 1992, SP-40(4), 901-913.[3]S.Z. Li, Markov Random Field Modeling in Computer Vision, Tokyo, Springer-Verlag, 1995.[4]Y.Y. Zhang, et al., Segmentation of Brain MR images through a hidden Markov random field model and the expectation maximization algorithm, IEEE Trans. on Medical Imaging, 2001,MI-20(1), 15-22. -
計(jì)量
- 文章訪問數(shù): 3105
- HTML全文瀏覽量: 148
- PDF下載量: 1152
- 被引次數(shù): 0
下載:
