融合图像质量评价指标的相关性分析及性能评估

Validation and Correlation Analysis of Metrics for Evaluating Performance of Image Fusion

图像融合质量评价指标研究旨在提供一种高效、准确的方法,为融合模型选择、参数优化等问题提供支持.本文通过对现有指标的机理分析、指标性能检验与指标间相关性分析,提出一种客观评价指标集的遴选策略.本文首先将现有客观评价指标归为三类:基于统计的、基于信息的和基于人类视觉系统的;之后列举了类别内经典指标和最新指标;并在标准数据集上,使用正确排序指标对各图像融合客观评价指标的性能进行验证.结果表明,基于视觉系统类的指标性能普遍优于前两类.最后,利用Spearman相关系数挖掘各指标间的相关程度.实验表明,通过指标性能和相关系数可以选取合适的客观评价指标集.

Image fusion performance evaluation aims at providing an efficient and accurate method for the fusion model choosing, parameter optimizing and the like. By analyzing the mechanism of existing metrics in theory and testing the performance of metrics and correlations with each other experimentally, the paper presents an effective metric set selection strategy. First of all, existing metrics are classified into three categories： statistics-based, information-based and human-visual-system based classes; secondly, we enumerate the classical or the latest metrics for each class. In addition, we test the performance of objective evaluating metrics in terms of correct ranking by running on a standard data set, and the results indicate that human-visual-system based metrics are superior to others. Finally, we explore correlations among metrics using Spearman correlation coefficient. Experimental results indicate that we can choose a proper objective evaluating metric set by means of performances and correlations of metrics.