FlexHDR: Modeling Alignment and Exposure Uncertainties for Flexible HDR Imaging.

Area of Science:

• Computer Vision
• Image Processing

Background:

• High dynamic range (HDR) imaging is crucial for digital photography, merging multiple low dynamic range (LDR) images to handle varying illumination.
• Common challenges include ghosting artifacts caused by misalignment and over-exposure due to motion and exposure uncertainties.

Purpose of the Study:

• To develop a new HDR imaging technique that explicitly models alignment and exposure uncertainties.
• To improve the quality of HDR images by robustly merging LDR frames and minimizing artifacts.

Main Methods:

• A novel strategy is introduced that jointly aligns images and assesses alignment/exposure reliability using an HDR-aware, uncertainty-driven attention map.
• A progressive, multi-stage image fusion approach is proposed for flexible, permutation-invariant merging of LDR images.

Main Results:

• The proposed method achieves up to 1.1dB PSNR improvement over state-of-the-art techniques.
• Subjective evaluations show enhanced details, improved colors, and a significant reduction in artifacts like ghosting.

Conclusions:

• The new HDR imaging technique effectively addresses alignment and exposure uncertainties, leading to superior HDR image quality.
• The uncertainty-driven attention map and multi-stage fusion approach offer a robust solution for generating high-quality HDR images.