LPIFM: A Learned Pairwise Preference Measure for Infrared-Visible Fusion Assessment
A new model called the Learned Perceptual Image Fusion Measure (LPIFM) has been developed by researchers to evaluate infrared-visible image fusion (IVIF) algorithms according to human preferences. Unlike conventional IVIF assessments that depend on scalar objective metrics—often misaligned with human evaluation—LPIFM utilizes a scalable surrogate based on the A/B/Tie comparison method. This model simultaneously analyzes two sources and two fused images to determine which is superior or if they are perceptually similar. Trained on a comprehensive preference dataset featuring all 6,300 unordered comparisons among 25 fusion algorithms, LPIFM effectively mitigates the quadratic expense of direct pairwise comparisons, enhancing the repeatability and scalability of subjective evaluations. This research is detailed in arXiv paper 2608.01301v3, available at https://arxiv.org/abs/2608.01301.
Key facts
- LPIFM is a source-conditioned model for ranking infrared-visible fusion algorithms.
- It predicts whether A is better, B is better, or the two are perceptually equivalent.
- The model is trained on a dense preference corpus covering all 6,300 unordered comparisons among 25 fusion algorithms.
- Traditional IVIF evaluation uses scalar objective metrics that often disagree with human judgment.
- Direct pairwise comparison is an established protocol but its cost grows quadratically with the number of algorithms.
- LPIFM operationalizes the human A/B/Tie comparison protocol as a repeatable, scalable surrogate.
- The paper is available on arXiv with ID 2608.01301v3.
- The announcement type is replace-cross.
Entities
—