Quality assessment of multi-exposure image fusion by synthesizing local and global intermediate references

Abstract

Multi-exposure fusion (MEF) takes a sequence of images with different exposure levels as input and generates a fused image that is more informative and perceptually appealing than any of the input images as output. During the past decades, many MEF algorithms have been proposed. Therefore, how to effectively compare the performance of different MEF algorithms is of great significance. Despite of this, research efforts on objective image quality assessment (IQA) of MEF images remain limited. In this paper, we propose a novel full-reference (FR) IQA method for MEF images by generating Local and Global Intermediate References (LGIR) from the input multiple images. Specifically, the intermediate reference features are synthesized in gradient domain, structural tensor domain, and global perception domain, respectively. The gradient and structure tensor domains reflect the local structural perception of the human visual system (HVS), while the global perception domain integrately considers the overall perception. In each domain, a single quality measure is estimated to reflect the visual quality of the fused image from a specific perspective. In addition, on considering the multi-scale property of the HVS, we estimate those quality measures at multiple scales, and fuse them together to predict the final quality score. Experimental results demonstrate the superiority of LGIR, achieving higher consistency with subjective quality scores than existing relevant FR-IQA methods.