纹理分析窗大小的高斯-马尔可夫随机场模型估计方法

GMRF model based window size estimation approach for texture analysis

在纹理分析中 ,窗口大小的选择对所提取特征的有效性及计算速度等有很大影响。文中利用高斯 -马尔可夫随机场 (GMRF)模型对纹理进行描述 ,采用最小平方误差估计获得纹理图像的随机场参数 ,并证明了这种估计的一致性。针对估计式在某些情况下可能无解 ,对该式作了改进 ,使其在实际应用中总能有解。利用估计的一致性 ,提出了一种系统估计纹理分析窗口大小的方法 ,实验表明了这种方法的有效性。

In texture analysis, the selection of window size has great influence on effectiveness of extracted feature and computing speed. In this paper, Gauss\|Markov random field model is employed to describe textures, the least square error approach is employed to estimate the field parameters, and the non\|bias feature of the estimation is proved. As there may be no solution according to this expression, an improvement is presented. Based on the non\|bias feature of parameter estimation, a window size estimation approach for texture primitives is presented, and experiment shows the effectiveness of our approach.