QoE-based multi-exposure fusion in hierarchical multivariate Gaussian CRF

Area of Science:

• Computer Vision
• Image Processing
• Perceptual Computing

Background:

• Existing multi-exposure image fusion methods focus on technical details but often neglect perceptual quality.
• Viewer experience in fused images is significantly impacted by factors like local contrast and color saturation.
• A gap exists in understanding and quantifying perceptual factors for improved image fusion.

Purpose of the Study:

• To introduce novel perceptual quality measures for multi-exposure image fusion.
• To develop a fusion method incorporating these perceptual factors.
• To enhance the viewer's quality of experience in fused images.

Main Methods:

• Proposed two perceptual quality measures: perceived local contrast and color saturation.
• Developed a hierarchical multivariate Gaussian conditional random field model.
• Embedded perceptual measures within the proposed conditional random field model for fusion.

Main Results:

• The proposed method demonstrates improved performance in multi-exposure image fusion.
• Generated fused images exhibit superior quality compared to existing state-of-the-art methods.
• The approach is effective across a variety of complex scenes.

Conclusions:

• Incorporating perceptual quality measures significantly enhances multi-exposure image fusion.
• The proposed hierarchical model effectively integrates local contrast and color saturation.
• This work provides a foundation for perceptually optimized image fusion techniques.