基于均值/方差分类的三维SOM初始化模式库算法

An initial pattern library algorithm based on mean/variance classification for 3D SOM

针对传统的初始化模式库算法存在模式矢量利用率低、与信源匹配程度不高的不足,提出一种基于均值/方差分类三维SOM初始化模式库算法。根据均值分类,将训练矢量集按照方差排序,以相同间隔抽取矢量组成初始化模式库。将算法运用到基于三维SOM算法的图像编码,结果表明,均值/方差分类初始化模式库算法无效模式矢量数量少、与信源匹配程度高,能有效地提高三维SOM算法的性能。

The pattern vector utilization and source matching degree is low for the traditional initial pattern library algorithm, so a new initial pattern library algorithm based on mean/variance classification for 3D SOM is proposed. The training vectors are sorted by mean value, then the training vectors in each part are sorted by variance and initial pattern library is chosen in pattern vectors at the same intervals. Experimental results show that the initial pattern library algorithm based on mean/variance classification has less invalid pattern vectors and high source matching degree. It is an effective way to improve the performance of 3D SOM algorithm.