Reconstructing Interlaced High-Dynamic-Range Video Using Joint Learning.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Sequential multi-exposure capture for High Dynamic Range (HDR) video suffers from ghosting artifacts due to motion.
• Interlaced exposure video imaging offers an alternative for HDR but introduces jaggy artifacts and sensor noise.
• Existing methods struggle to balance dynamic range extension with image quality in video.

Purpose of the Study:

• To propose a data-driven approach for jointly solving deinterlacing and denoising problems in interlaced exposure HDR video.
• To reconstruct detailed extended dynamic range information from interlaced video inputs.
• To enhance image quality in HDR video by mitigating artifacts associated with interlaced exposures.

Main Methods:

• Joint dictionary learning via sparse coding is employed for the deinterlacing problem, leveraging partial information from differently exposed rows.
• Sparse coding is tailored to address additive noise in low- and high-exposure rows for denoising.
• Multiscale homography flow is adopted for temporal sequence denoising.

Main Results:

• The proposed method effectively reconstructs details for extended dynamic range video from interlaced inputs.
• Joint deinterlacing and denoising significantly improve image quality compared to traditional methods.
• Demonstrated advantages over state-of-the-art high-dynamic-range video techniques.

Conclusions:

• The developed data-driven approach enables concurrent capture of HDR video frames without ghosting artifacts.
• This method offers a robust solution for deinterlacing and denoising in interlaced HDR video imaging.
• The technique shows significant potential for advancing HDR video technology by improving image fidelity and quality.