Improving the perception of low-light enhanced images

Area of Science:

• Computational imaging
• Computer vision
• Image processing

Background:

• Low-light image enhancement is crucial for applications like surveillance and autonomous driving.
• Current methods often optimize distortion metrics (e.g., Peak Signal-to-Noise Ratio, Structural Similarity Index Measure), which do not guarantee perceptual quality.
• The perception-distortion trade-off highlights that minimizing distortion can negatively impact perceived image quality.

Purpose of the Study:

• To develop a post-processing method for low-light images that enhances perceptual quality.
• To achieve results similar to existing methods but with improved visual perception.
• To address the limitations of distortion-metric-focused enhancement techniques.

Main Methods:

• A post-processing approach is proposed, taking a low-light image and an existing enhanced image as input.
• The method hypothesizes that minimal perceptual modification involves local shading adjustments and global illumination color changes.
• The technique aims to preserve the essence of the original enhancement while boosting perceptual quality.

Main Results:

• Quantitative evaluation using perceptual blind image quality assessment metrics (e.g., BRISQUE, NIQE, UNIQUE) demonstrated significant improvements.
• User preference tests confirmed the enhanced perceptual quality of the processed images.
• The method successfully improved perceived image quality without drastically altering the output of existing enhancement algorithms.

Conclusions:

• The proposed post-processing method effectively enhances the perceptual quality of low-light images.
• It offers a valuable alternative to traditional methods by focusing on perceptual improvements rather than solely distortion minimization.
• This approach contributes to the field of computational color imaging by providing better visual results for low-light scenarios.