Edge-Enhanced with Feedback Attention Network for Image Super-Resolution.

Area of Science:

• Computer Vision
• Deep Learning
• Image Processing

Background:

• Single Image Super-Resolution (SISR) is crucial for enhancing low-resolution images.
• Deep Convolutional Neural Networks (CNNs) have advanced SISR, but existing methods suffer from artifacts and edge blurring.
• Attention and feedback mechanisms offer potential for improved feature utilization and output refinement in SISR.

Purpose of the Study:

• To propose an effective deep learning model for SISR that addresses artifacts and edge blurring.
• To integrate attention and feedback mechanisms for enhanced feature learning and refinement.
• To achieve state-of-the-art SISR performance with reduced computational complexity.

Main Methods:

• Developed an Edge-enhanced Feedback Attention Network (EFANSR) for SISR.
• Incorporated channel and spatial attention blocks for adaptive feature learning.
• Implemented a feedback mechanism for high-level information integration and input refinement.
• Designed an edge enhancement network to sharpen image edges.

Main Results:

• The proposed EFANSR method effectively reduces artifacts and edge blurring in super-resolved images.
• The network adaptively learns features using integrated attention mechanisms.
• The feedback mechanism contributes to fine-tuning the super-resolution process.
• Experimental results demonstrate comparable performance to state-of-the-art methods with lower complexity.

Conclusions:

• The EFANSR model offers a novel approach to single image super-resolution.
• Integrating feedback and attention mechanisms significantly improves image quality.
• The method provides a promising solution for high-quality SISR with computational efficiency.