Using ROC curves and AUC to evaluate performance of no-reference image fusion metrics

Abstract

Image fusion has many applications in which a reference image is not always available including image registration, medical imaging, and fusion between visible and infrared imagery. For these no-reference applications, it is important that there are objective and efficient methods for validating fusion performance, as subjective image fusion evaluation is time consuming and non-scalable. There have been multiple no-reference objective metrics created in the past. These include mutual information, spatial frequency, and structural similarity index measure (SSIM). However, it is important to consider justification of a given evaluation metric as appropriate for a given type of image fusion method. We seek to ensure that if a given metric scores one image higher than another, then the image with the higher metric score is subjectively preferred. This pilot study investigates the applications of Receiver Operating Characteristic (ROC) curves and the Area Under the Curve (AUC) as a method of validation for fusion metrics used for evaluating image fusion methods. The results from the pilot study indicate that ROC curves and AUC provide a discriminating form of validation for image fusion metrics to support image fusion applications evaluation.