纹理数据的可分离性评价指标研究

Research on Separability Index for Gray Co-occurrence Matrix Texture Data

根据灰度共生矩阵计算的纹理数据,受到窗口大小、步长、方向、灰度等级这些参数及纹理统计量等因素的影响。如何评价在计算机自动搜素过程中,不同参数组合及不同纹理统计量的纹理数据,是一个值得研究的问题。纹理数据如果具有好的区分度即好的图像分类价值,则在图像同一地物区域内,纹理数据应该具有较好的一致性,而不同地物区域间纹理数据应该具有明显的差异。根据这个原则,本文提出了一个可分离性指标——J指标。无人机航片的试验结果表明,该指标具有较好的评价效果。试验同时表明:窗口大小至少应包含图像中纹理颗粒最大地物的一个周期;总体来说,J值随着窗口增大而增大;森林纹理与方向关系不明显。

Texture data based on the grey co-occurrence matrix technique are often subject to parameters such as window size,step size,direction,grey scale as well as different statistics.How to evaluate the separability of the texture data in image segmentation is an important job.If one suite of texture data has a good separability,then the data should have a high homogeneity in the same region and be different between regions.Upon this assumption this paper proposed an index,J index for evaluating the separability of texture data.Examples using unmanned aerial vehicle images give us a good result about this index.Examples also show that windows should at least contain one whole texture cycle of a ground cover that may has maximum texture cycles in the image,J value gets larger when the window becomes larger,and the forest texture has very low relationship with direction.