Multiple-Exposure Image Fusion for HDR Image Synthesis Using Learned Analysis Transformations.

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Modern imaging demands High Dynamic Range (HDR) capabilities.
• Low-cost sensors have limited dynamic range, hindering HDR imaging.
• Multiple-exposure image synthesis offers a viable solution for low-cost HDR imaging.

Purpose of the Study:

• To develop an effective image-fusion algorithm for synthesizing HDR images from multiple exposures captured by low-cost sensors.
• To combine established fusion methods to improve HDR image quality and reduce artifacts.

Main Methods:

• A hybrid image-fusion approach combining two distinct methods.
• Fusion of the luminance channel using the Mitianoudis and Stathaki (2008) method.
• Combination of color channels utilizing the Mertens et al. (2007) method.

Main Results:

• The proposed fusion algorithm successfully generates high dynamic range images.
• The method effectively avoids halo artifacts prevalent in other state-of-the-art techniques.
• Extended analysis and experimental results validate the method's effectiveness.

Conclusions:

• The combined fusion strategy provides a robust solution for low-cost HDR imaging.
• The developed algorithm achieves superior image quality without common fusion artifacts.
• This work represents an enhanced version of prior conference research, offering deeper insights and validation.