A Perceptual Quality Metric for Performance Evaluation of Image Fusion

Usually, comparative evaluation by human visual inspection in image fusion is often used to assess the relative fusion performance of different fusion schemes. In this paper, a novel non-reference and quantitative quality metric is proposed to evaluate the performance of the fusion algorithms objectively. Instead of using traditional error summation methods, the proposed measure is designed to reflect humanpsilas visual perception and requires no knowledge of the ground truth. The proposed measure is based on wavelet tranform, which is easy to calculate and can implement straight-forwardly to assess the quality of the fusion image. Some commonly used fusion algorithms are performed to evaluate the effectiveness of the proposed metric. Experiments on various image types show that the measure has a significint improvement over standard measures, such as root mean squared error (RMSE) and mutual informance(MI), and coincides with subjective human visual inspection well.