ROC curve analysis for validating objective image fusion metrics

Image fusion is a process that allows for the synthesis of information from multiple source images into a single image. There are many applications for image fusion including night vision, medical imaging, and remote sensing. Over the many applications, numerous image fusion algorithms have been explored from averaging pixel intensities to fusion through multi-resolution decomposition transforms such as the wavelet or contourlet. Objective evaluation of a given image fusion method is still a major challenge especially when there exists no reference image. Existing no-reference objective fusion metrics include information theory based metrics, image feature based metrics, and structural similarity metrics. However there has been very little work done in validating which objective metric best evaluates a given image fusion algorithm. Receiver operating characteristic (ROC) curves and the area under the curve (AUC) provide a viable validation method for metric selection. This study focuses on validating objective fusion metrics over mutual information, spatial frequency, and structural similarity Index Measure (SSIM) used to evaluate fusion algorithms for denoising applications.