Original title: MEF-TBN: Multi-exposure image fusion approach based on a three-branch network for feature extraction and color enhancement
Chunmeng Wang, Wenxiang Zhang, Quan Zhang, Siyi Zhou · September 2026
Cameras struggle to capture very bright and very dark spots in one shot. Multi-exposure fusion combines several pictures taken at different brightness levels into a single balanced picture. However, older techniques often lose fine details, broad image context, or true colors. To fix this, researchers built MEF-TBN, a system with three specialized processing branches. The first branch studies small local features. The second branch looks across the whole image to connect distant parts. These two branches map out brightness information, which a merger module refines and sharpens using guided filtering. Meanwhile, the third branch focuses on color, learning how brightness and color shades link together to restore natural tones. When tested across three datasets, this approach beat other methods at keeping features and matching human perception. It also showed clear gains over a leading rival called SwinFusion.
You're merging three photos of the same scene on your phone: one too bright, one too dark, one just right. If you just overlay them, the bright parts wash out and the dark parts stay muddy. Here, the algorithm uses three separate helpers: one zooms into corners, one looks at the whole picture, and one adjusts colors. They each write notes on which parts to keep, then merge the notes into one clean image.
Multi-exposure fusion decides how your phone and camera merge shots taken at different brightness. Poor fusion loses sharp details or makes colors look fake. This network keeps both, and tests show it beats existing methods on metrics that match what human eyes perceive, meaning sharper, more natural-looking photos and clearer medical images.
The abstract doesn't state explicit limitations, but it reports ranking second on MEF-SSIM and testing on only three datasets, so broad applicability and structural similarity performance remain open.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
MEF-TBN: Multi-exposure image fusion approach based on a three-branch network for feature extraction and color enhancement — Chunmeng Wang, Wenxiang Zhang, Quan Zhang et al. — PLoS ONE — doi:10.1371/journal.pone.0357156
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.