图像分析的马尔可夫模型.pdfVIP

Mo deling Image Analysis Problems Using Markov Random Fields Stan Z. Li Microsoft Research China Beijing Sigma Center, Beijing 100080, China 1 Intro duction Mo deling problems in this article are addressed mainly from the computational viewp oint. The primary concerns are how to dene an ob jective function for the optimal solution for an image analysis problem and how to nd the optimal solution. The reason for dening the solution in an optimization sense is due to various uncertainties in imaging pro cesses. It may b e di脝cult to nd the p erfect solution, so we usually lo ok for an optimal one in the sense that an ob jective, into which constraints are enco ded, is optimized. Contextual constraints are ultimately necessary in the interpretation of visual information. A scene is understo o d in the spatial and visual context of the ob jects in it; the ob jects are recognized in the context of ob ject features at a lower level representation; the ob ject features are identied based on the context of primitives at an even lower level; and the primitives are extracted in the context of image pixels at the lowest level of abstraction. The use of contextual constraints is indisp ensable for a capable image analysis system. Markov random eld (MRF) theory provides a convenient and consistent way for mo deling context dep endent entities such as image pixels and correlated features. This is achieved through characterizing mutual in uences among such entities using conditional MRF distributions. The practical use of MRF mo dels is largely ascrib ed to a theorem stating the equivalence b etween MRFs and Gibbs distributions which was established by Hammersley