一种分层马尔可夫图像模型及其推导算法

离散分层马尔可夫随机场(MRF)模型由于层间具有了因果性,因而其非迭代的推导算法比非因果的马尔可夫随机场模型的迭代算法复杂度低得多,结果更精确.针对图像分割问题中观测数据有限的情况,提出了一种新的基于离散分层MRF的半树模型,推导出了它的最大后验边缘概率(MPM)算法.半树模型不仅继承了一般分层模型快速、误分类少的优点,还避免了计算中遇到的数值下溢问题,减轻了分层模型带来的块现象,尤其适合大幅面图像的处理.

A Hierarchical Markov Image Model and Its Inference Algorithm

The noniterative algorithm of discrete hierarchical MRF (Markov random field) model has much lower computing complexity and better result than its iterative counterpart of noncausal MRF model, since it has causality property between layers. A new model based on the hierarchical MRFhalf tree model is proposed for only one image can be obtained in image segmentation, whose MPM (maximizer of the posterior marginals) algorithm is inferred too. The proposed model not only inherits the advantages of general hierarchical MRF model but also does better: it makes large image more tractable within much less time, prevents data underflow appeared in computing, and alleviates the block artifacts occurred in hierarchical models. It is especially fit for large scale images.