A lightweight hybrid perception enhancement network for infrared image super-resolution

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Infrared image super-resolution (SR) is challenging due to limitations in current Convolutional Neural Network (CNN) and Transformer models.
• CNNs struggle with long-range dependencies, while Transformers are computationally intensive and miss local details.

Purpose of the Study:

• To develop a novel Hybrid Perception Enhancement Network (HPEN) for improved infrared image super-resolution.
• To address the limitations of existing SR methods by combining global context modeling and local detail extraction.

Main Methods:

• Proposes a Hybrid Perception Enhancement Network (HPEN) featuring a Hybrid Perception Enhancement Block (HPEB).
• HPEB integrates a Token Aggregation Block (TAB) for global context, a Multi-scale Feature Enhancement Block (MFEB) for local details, and a convolutional layer for refinement.

Main Results:

• HPEN achieves leading performance in infrared image super-resolution tasks, outperforming existing methods.
• Achieves the best Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) for the [Formula: see text] SR task among lightweight approaches.
• Demonstrates significant efficiency advantages, reducing FLOPs by 42.9% and GPU memory usage by 90.6% compared to HiT-SR, with a [Formula: see text] faster inference speed.

Conclusions:

• The proposed HPEN effectively models both global context and local details for superior infrared SR.
• HPEN offers a computationally efficient and high-performing solution for infrared image super-resolution.
• The developed network provides a significant advancement in lightweight SR approaches for infrared imaging.