A Lightweight Image Super-Resolution Reconstruction Algorithm Based on the Residual Feature Distillation Mechanism

Area of Science:

• Computer Vision
• Artificial Intelligence
• Deep Learning

Background:

• Convolutional Neural Networks (CNNs) are widely used for image super-resolution (SR).
• Increasing CNN depth does not always improve SR performance and can complicate training.
• Novel approaches are needed to enhance SR reconstruction without excessive network complexity.

Purpose of the Study:

• To propose a lightweight image super-resolution reconstruction algorithm (SISR-RFDM).
• To improve the recovery of high-frequency details like edges and textures.
• To enhance feature reuse and inter-layer information flow in SR.

Main Methods:

• Developed a Residual Feature Distillation Mechanism (RFDM) for lightweight SR.
• Incorporated Spatial Attention (SA) modules to guide the recovery of fine details.
• Utilized Global Feature Fusion (GFF) for hierarchical feature integration and reuse.

Main Results:

• The proposed SISR-RFDM algorithm demonstrated superior performance over comparative methods.
• Achieved a 0.23 dB improvement in Peak Signal-to-Noise Ratio (PSNR).
• Reached a Structural Similarity Index (SSIM) of 0.9607, indicating enhanced image quality.

Conclusions:

• The SISR-RFDM algorithm effectively reconstructs high-quality images using a lightweight architecture.
• The integration of RFDM, SA, and GFF significantly improves SR performance.
• This approach offers a promising solution for efficient and effective image super-resolution.