论文部分内容阅读
阐述了高斯-马尔可夫随机场模型的基本原理,建立了木材表面纹理的2-5阶高斯-马尔可夫随机场(Gauss-MRF)模型,用最小二乘法估计了300个木材样本表面纹理的2-5阶Gauss-MRF参数。数据分析表明,各不同纹理特征参数之间具有明显的分布性;Gauss-MRF参数值最大的参数所表示的纹理集聚方向为纹理的主方向;对于纹理主方向相同的样本,纹理越细致,其相应参数越大,而其他参数越小;Gauss-MRF阶数越高,纹理描述越细致;在2阶Gauss-MRF模型情况下,弦切纹理的B1参数大于径切纹理的B1;弦切纹理的B2、B3、B4分别小于径切纹理的B2、B3、B4。根据分离判据的值,确定以5阶Gauss-MRF参数为特征向量进行初步聚类,总体正确率为88%。
The basic principle of Gaussian-Markov random field model is expounded. A 2-5-order Gauss-Markov random field (Gauss-MRF) model of wood surface texture is established. The surface texture of 300 wood samples 2-5 Gauss-MRF parameters. The data analysis shows that there are obvious distributions between the different texture feature parameters. The direction of texture aggregation represented by the parameter with the largest Gauss-MRF value is the main direction of the texture. For the samples with the same main direction of texture, the texture is more detailed, The larger the corresponding parameters are, the smaller the other parameters are. The higher the Gauss-MRF order is, the more detailed the texture description. In the case of the second-order Gauss-MRF model, the B1 parameter of the chord cut texture is larger than the diameter cut B1; B2, B3, B4 are smaller than the diameter cut texture of B2, B3, B4. According to the value of the separation criterion, the initial clustering with the fifth-order Gauss-MRF parameters as the eigenvector is determined, and the overall correctness rate is 88%.