Unsupervised Image Segmentation Based on Incomplete Hierarchical MRF
Xi Wang · Dianzi xuebao · 2004
Hierarchical MRF image model has causality property between layers,and the causality is consistent with the characteristics of images.So the processing time of such models is much less than that of the plain MRF models.An expectation maximization(EM) algorithm estimating parameters of incomplete hierarchical MRF model,which is a new hierarchical MRF image model we presented,is deduced.The advantage of less time cost possessed by the non iterative algorithm of hierarchical models is inherited.Time is further reduced due to simplified model structure.The interaction between neighbor nodes on the top layer is considered,which results in more accurate estimate values with less computing cost.The algorithm is used in unsupervised image segmentation.The experimental results demonstrate that it is characterized by high speed and better results compared with that of hierarchical models.It is more fit for large images.