Abstract: One of the main problems related to global motion estimation is that the algorithm only considers the change information in the difference image, but the import information of segmentation and correlation has been ignored. To overcome such drawbacks, a technique of global motion estimation based on MAP is proposed. The system energy is defined through MAP and the potential energy is defined by segmentation mask and the prior information. Using the system energy as object function, the global motion is estimated. Experimental results confirm the effectiveness of the proposed technique.

Key words: global motion estimation, MAP, Gibbs energy

摘要： 常见的全局运动算法存在的问题是仅仅考虑像素的残差，而忽略了相邻像素和分割信息等因素对算法的影响。为了解决上述问题，该文提出一个基于最大后验概率(MAP)的全局运动估计算法。通过MAP算法定义系统能量，引入先验能量来描述像素的相邻关系和分割信息，将系统能量作为目标函数通过LM算法估计全局运动参数。实验结果证明了该算法的有效性。

关键词: 全局运动估计, 最大后验概率, Gibbs能量

CLC Number: 

• TP18