一种改进的网格数据保持梯度滤波方法

 基于网格数据滤波的基本思想，针对杨高印提出的小子域滤波法存在子域划分重心不稳的问题，本文提出了新的子域划分方法，即通过考察滤波窗口内数据的数值分布或梯度变化情况，在充分考虑子域划分对称性的基础上，采用对称子域的均方根误差或平均梯度大小差值为判别指标确定数据变化的主梯度方向，并用此判别指标代替数据变化平缓系数判别指标，在滤波输出的取值上考虑数据变化的梯度方向上的两个相对称子域对输出的贡献大小进行加权平均，使滤波输出结果更稳定。通过数据试算，表明网格数据滤波法输出结果更加合理、稳健。

An improved gradient-preseved grid data filtering approach

Based on the basic idea of grid data filtering and in view of the issue of unstable center of gravity in subdomain partitioning that exists in subdomain filtering approach, presented by Mr.Yand Gao-yin,the paper presented a new subdomain partitioning approach, that is: through observing the numeric distribution or gradient variation of data in filtering window and on the basis of fully considering the symmetry of subdomain partitioning,adopting the mean-square-root error or the size of average gradient in symmetric subdomain as distinguish index to determine the main gradient direction of data variation and using this distinguish index to replace the distinguish index of data variation smoothing coefficient;as concerns the taking number of filtering output,the implement of weighted mean should consider the size of contribution of two symmetric subdoamins in gradient direction of data variation to filtering output,making the results of filtering output more stable.The results of data test showed the output results of grid data filtering are more reasonable and robust.