Pairwise Operator Learning for Patch-Based Single-Image Super-Resolution

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Single-image super-resolution (SISR) aims to reconstruct high-resolution (HR) images from low-resolution (LR) inputs.
• Existing SISR methods often require significant computational resources and storage.
• Representing image patches as matrices offers a novel perspective for SISR algorithm development.

Purpose of the Study:

• To propose a novel patch-based regression algorithm for single-image super-resolution.
• To leverage matrix space representation for learning regression operators in SISR.
• To develop an efficient and effective SISR method with reduced data storage requirements.

Main Methods:

• Image patches are treated as matrices for SISR.
• Regression operators are learned in a matrix space to map LR to HR patches.
• Pairwise operators (left and right multiplication) extract row and column information from LR patches.

Main Results:

• The proposed patch-based regression algorithm demonstrates efficiency in both training and testing phases.
• The algorithm requires significantly less data storage compared to popular SISR methods.
• Experimental results confirm competitive super-resolution performance, comparable to existing state-of-the-art algorithms.

Conclusions:

• The matrix-based approach provides an efficient and effective solution for single-image super-resolution.
• Treating image patches as matrices simplifies the learning of regression operators.
• The proposed method offers a promising alternative for practical SISR applications due to its efficiency and performance.