Super-resolution reconstruction based on Gaussian transform and attention mechanism.

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Image super-resolution (SR) aims to enhance low-resolution (LR) images.
• Existing methods often struggle with recovering fine details and high-frequency information.

Purpose of the Study:

• To develop a novel super-resolution reconstruction network for improved image detail and texture recovery.
• To enhance the performance of super-resolution reconstruction using advanced deep learning techniques.

Main Methods:

• A new super-resolution reconstruction network incorporating multi-scale Gaussian difference transform, attention mechanism, and feedback mechanism.
• Utilizing both pixel loss and texture loss functions to focus on structure and texture learning.
• Increasing network depth to better capture high-frequency features.

Main Results:

• The proposed method significantly strengthens details in low-resolution blurred images.
• Enhanced expression of high-frequency features through attention mechanisms and increased network depth.
• Superior performance compared to existing methods in quantitative and qualitative evaluations.

Conclusions:

• The novel network effectively recovers high-frequency detail information in super-resolution reconstruction.
• The combined approach of multi-scale transforms, attention, and dual loss functions leads to superior image restoration.
• This method advances the field of image super-resolution by improving detail and texture fidelity.