双色中波红外图像差异纹理特征分析方法研究

The Analysis Method of Differences in Image Texture Features Based on Dual-color MWIR Images

针对双色中波红外图像特征空间维数高,不易于对其双波段成像差异进行综合分析的问题,提出了一种基于赋权思想及雷达图的双波段图像差异纹理特征的分析方法。该方法首先构造一个特征选择矩阵,利用其选择出能够满足双波段图像差异分布规律的有效纹理特征;然后利用赋权思想将差异不显著的纹理特征去除;最后采用雷达图对保留下的差异特征进行高维显示,并由雷达图中多边形的形状信息得到维数较少的图形特征。实验数据表明,降维后的特征能够综合反映双波段图像的纹理差异幅度,为后期差异特征驱动的多级融合方法探索奠定了基础。

In order to overcome the difficulty that feature space dimension of the dual-color MWIR infrared image is large and hard to analyze synthetically for its imaging differences, a new analytical method based on the empowerment thinking and radar chart concerned with differences in texture features of dual-band image is presented. The method firstly uses a feature selection matrix to select the effective texture features which can meet the distribution rules of different features in dual-band image. Then the empowerment thinking is used to remove the texture features without significant changes. Finally, the method adopts radar chart to display preserving differential features in a high-dimensional, further getting the graphic features with fewer dimensions by polygon information of radar charts. According to the experiment result, the low-dimensional features can reflect the magnitude of dual-band image texture differences, laid the foundation of exploring multi-level fusion methods driven by different features.