Method for reconstructing a high dynamic range image based on a single-shot filtered low dynamic range image.

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Traditional cameras struggle with high dynamic range (HDR) imaging due to sensor limitations, often requiring multiple exposures.
• Existing methods for single-shot HDR capture are insufficient for complex scenes.

Purpose of the Study:

• To develop an improved single-shot method for reconstructing high dynamic range (HDR) images.
• To overcome the limitations of traditional cameras in capturing images with a wide range of brightness levels.

Main Methods:

• A novel approach using a single-exposure filtered low dynamic range (FLDR) image captured with an optical filter.
• Development of a deep inverse tone mapping network (DITMnet) featuring multibranch feature extraction and multioutput image synthesis.
• Reconstruction of HDR images from the single FLDR image.

Main Results:

• The proposed method successfully reconstructs HDR images from a single FLDR image.
• Experimental results demonstrate superior performance compared to similar algorithms across various exposure states and color spaces.
• The DITMnet effectively handles different RGB channel exposure states and luminance ranges.

Conclusions:

• The developed single-shot HDR image reconstruction method using FLDR images is effective and outperforms existing techniques.
• The integration of optical filtering and deep learning offers a promising solution for capturing challenging HDR scenes.
• This approach advances the capabilities of single-shot imaging for high dynamic range applications.