FRESH-FRI-Based Single-Image Super-Resolution Algorithm.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Single image super-resolution (SISR) is crucial for enhancing image detail.
• Existing SISR methods often require large datasets of paired low- and high-resolution images.
• A need exists for data-driven, efficient SISR algorithms.

Purpose of the Study:

• To propose a novel single image super-resolution algorithm.
• To achieve state-of-the-art performance without relying on external patch pairs.
• To enhance resolution by modeling image lines as piecewise smooth functions.

Main Methods:

• Modeling image lines as piecewise smooth functions.
• Utilizing the theory of sampling signals with finite rate of innovation (FRI).
• Combining FRI reconstruction with traditional linear methods via wavelet-based multi-resolution analysis and filter banks.
• Applying the fused reconstruction along vertical, horizontal, and diagonal image directions.
• Incorporating error learning from lower resolution results for further improvement.

Main Results:

• The proposed method achieves superior performance compared to state-of-the-art algorithms.
• Outperformance is demonstrated across various blurring kernels.
• The algorithm effectively enhances resolution without external training data.

Conclusions:

• The novel approach offers a data-free solution for single image super-resolution.
• Modeling image lines with FRI and wavelet theory provides a robust reconstruction framework.
• The method presents a significant advancement in image resolution enhancement techniques.