Edge-enhanced infrared image super-resolution reconstruction model under transformer.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Low resolution of infrared images hinders applications in military, security, and surveillance.
• Existing super-resolution methods struggle with edge information recovery and ringing artifacts.

Purpose of the Study:

• To propose an edge-enhanced infrared image super-resolution reconstruction model (TESR) using Transformer architecture.
• To address the challenges of edge detail loss and ringing effects in infrared image super-resolution.

Main Methods:

• An edge detection auxiliary network is integrated to enhance edge information from low-resolution inputs.
• CSWin Transformer is employed for parallel self-attention computation, expanding the receptive field and utilizing higher-level semantic features.

Main Results:

• The TESR model effectively extracts comprehensive image information and more accurate edge details.
• Enhanced texture details and improved super-resolution reconstruction results are achieved for infrared images.

Conclusions:

• The proposed TESR model offers a significant advancement in infrared image super-resolution.
• Accurate edge information enhancement and artifact reduction are key contributions for improved image quality.