一种基于MRF理论去除Range图象噪声的方法

A METHOD FOR REDUCING NOISE IN RANGE DATA IMAGES BASED ON MRF THEOREM

Ｒａｎｇｅ图象的相关结构可被认为是受噪声干扰的分段光滑面．一旦曲面的参数被确定，那么就可以利用这些参数重建曲面，从而大大减少图象中的噪声．在Ｒａｎｇｅ图象中，几乎所有象素的统计特性往往与其邻近象素的统计特性相关．文中利用ＭａｒｋｏｖＲａｎｄｏｍＦｉｅｌｄ（ＭＲＦ）理论来模拟这种相关性，将曲面参数确定问题转化为一个后验均值求解问题．

The relevant structure of a Range data image is viewed as a piecewise smooth surface contaminated by noise.Once the surface parameters are determined,the Range data image can be reconstructed by using the parameters to get new pixel values,and consequently more noise in the Range data image can be reduced.In Range data images,almost all pixel characteristics are statistically and geometrically correlated with neighboring pixel characteristics.In this approach,the surface parameters are modeled as MRF on 2D lattices and the surface parameters determination problem is then formulated as a rule to find the posteriori mean estimation.