Unified Single-Image and Video Super-Resolution via Denoising Algorithms

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Single Image Super-Resolution (SISR) and Video Super-Resolution (VSR) are crucial for enhancing image quality.
• Traditional VSR methods face challenges with motion estimation and video dynamics, limiting direct extension from SISR.
• Existing denoising algorithms are highly effective for image and video noise reduction.

Purpose of the Study:

• To propose a unified and robust framework for both SISR and VSR.
• To adapt effective denoising techniques for super-resolution tasks.
• To develop a VSR algorithm that bypasses the need for explicit motion estimation.

Main Methods:

• Utilizing the plug-and-play-prior framework and regularization-by-denoising (RED).
• Applying existing image and video denoisers within a unified super-resolution formulation.
• Harnessing the VBM3D video denoiser for VSR.

Main Results:

• Demonstrated a unified framework applicable to both SISR and VSR.
• Achieved a competitive motion-estimation-free VSR algorithm.
• Showcased high-quality output and fast processing in VSR.

Conclusions:

• The proposed framework effectively addresses SISR and VSR challenges by leveraging denoising priors.
• This approach offers a simpler and more robust alternative to traditional VSR methods.
• The unified formulation highlights the potential of denoising techniques in solving complex inverse problems like super-resolution.